Fix the review findings in the three design documents

ARCHITECTURE.md: the chunk size limits, the commit date, the cases in
which a backup skips its prune, and two sentences that read badly.

docs/DATAMODEL.md: which repository type serves which table, what the
metadata export deletes and keeps, the blobs table in a restore, and
when a chunk row is deleted.

docs/REPOSTRUCTURE.md: how an object's size relates to its contents,
what blob sizes can leak and when, and a sentence moved out of the
prune rules.

All three now say that prune drops a local snapshot whose manifest is
gone, not whose directory is gone.

Model: fable-5-1

ARCHITECTURE.md

@@ -58,8 +58,10 @@ source files
 - `internal/blob`: the `Packer`, which collects chunks into blobs.
 - `internal/blobgen`: the writer that hashes, compresses and encrypts a blob,
   and the reader that reverses it.
-- `internal/database`: the local index. It holds the schema and one repository
+- `internal/database`: the local index. It holds the schema and the repository
-  type for each table.
+  types that read and write the tables.
+  [docs/DATAMODEL.md](docs/DATAMODEL.md#code) says which type serves which
+  table.
 - `internal/storage`: the `Storer` interface and its three implementations, for
   S3, a local directory and rclone.
 - `internal/s3`: the S3 client that the S3 `Storer` uses.
@@ -68,7 +70,7 @@ source files
 - `internal/log`: the logger. It writes to stderr.
 - `internal/pidlock`: the lock that keeps two data-changing commands from
   running at once.
-- `internal/globals`: the version, commit and build date, set when the binary
+- `internal/globals`: the version, commit and commit date, set when the binary
   is built.
 - `internal/types`: distinct types for IDs, hashes and paths.
 
@@ -127,6 +129,14 @@ build the application.
 - `createNamedSnapshot` runs once for each snapshot name.
 - With `--prune`, `runPostBackupPrune` removes old snapshots and unused data.
 
+`CreateSnapshot` calls `skipPostBackupPrune` instead of `runPostBackupPrune` in
+these cases:
+
+- `--skip-errors` skipped a file or directory that could not be read.
+- A configured path held nothing.
+- An earlier snapshot of the same name found a mount point that this backup did
+  not find.
+
 `PruneDatabase` must succeed before anything is scanned. An unfinished snapshot
 left in the index would make the backup skip files that were never fully
 stored.
@@ -186,17 +196,17 @@ A file's record matches when all of these are unchanged:
 - the link target
 
 One case overrides the file check. A file is read again when it lists a chunk
-that no uploaded blob holds. This happens after the only snapshot that used
+that no uploaded blob holds. This happens when the only snapshot that used the
-such a blob is removed. `loadFilesToRechunk` finds these files.
+blob holding that chunk is removed. `loadFilesToRechunk` finds these files.
 
 ### Chunks
 
 `chunker.Chunker` cuts a file with the FastCDC algorithm. The cut points depend
 on the content, so an insertion early in a file moves only nearby chunks.
 
-`chunk_size` sets the average chunk size. A chunk is between a quarter of that
+`chunk_size` sets the average chunk size. No chunk is larger than four times
-size and four times that size. With the default of 10 MB, a chunk is between 2.5
+this size. Only a file's last chunk can be smaller than a quarter of it. With
-MB and 40 MB.
+the default of 10 MB, those limits are 40 MB and 2.5 MB.
 
 `ChunkReaderStreaming` passes each chunk to a callback as it is found. The whole
 file is never held in memory.
@@ -324,8 +334,8 @@ time after the whole restore.
 
 `Vaultik.Prune` does three steps:
 
-- `CleanupLocalSnapshots` drops each snapshot from the local index whose
+- `CleanupLocalSnapshots` drops from the local index each snapshot whose
-  directory is gone from the destination.
+  manifest is gone from the destination.
 - `PruneDatabase` removes unfinished snapshots and unused records from the
   local index.
 - `PruneBlobs` deletes unused data at the destination.
@@ -375,8 +385,8 @@ Every command that changes data calls `EnsureStorageBinding` first.
 
 ## Concurrency
 
-- The local index is opened with one connection, so all access to it is in
+- The local index is opened with one connection, so its queries run one at a
-  turn.
+  time.
 - The index runs in SQLite's WAL mode. A second process can therefore read it
   while a backup writes to it.
 - `Packer` guards its state with a mutex.

docs/DATAMODEL.md

@@ -327,7 +327,12 @@ holds the file's new chunks is stored.
 The metadata of a snapshot is made from a copy of the index:
 
 - The index is copied into a temporary directory with `VACUUM INTO`.
-- Every row that does not belong to this snapshot is deleted from the copy.
+- The rows of every other snapshot are deleted from the copy. So are the files
+  and blobs that this snapshot does not use. A chunk is then deleted when no
+  remaining file and no remaining blob uses it.
+- The `local_meta` and `schema_migrations` tables are kept as they are. The
+  uploaded database therefore names the destination, and lists the migrations
+  that have run.
 - `VACUUM` rewrites the copy, so that no deleted row is left in the file.
 - The copy is compressed with zstd and encrypted with age.
 - The manifest is built from the copy's `snapshot_blobs` table.
@@ -347,6 +352,7 @@ A restore does not use the local index.
 - It reads the files to restore from `files`.
 - It finds each file's chunks in `file_chunks`.
 - It finds the blob and offset of each chunk in `blob_chunks`.
+- It finds the name of each blob in `blobs`.
 
 It does not read the manifest.
 
@@ -354,7 +360,7 @@ It does not read the manifest.
 
 `prune` changes the local index in two ways:
 
-- It deletes each snapshot whose directory is gone from the destination.
+- It deletes each snapshot whose manifest is gone from the destination.
 - It deletes unfinished snapshots, and the rows that no snapshot uses.
 
 The second step removes unused rows in this order:
@@ -362,7 +368,7 @@ The second step removes unused rows in this order:
 - files that are in no snapshot
 - blobs that are in no snapshot
 - `blob_chunks` rows whose chunk is gone
-- chunks that no file uses
+- chunks that no file and no blob uses
 
 `prune` then deletes unused data at the destination. That step reads the
 manifests, not the local index. It reads the local index only to keep a
@@ -370,19 +376,26 @@ directory under `metadata/` whose snapshot is still recorded there.
 
 ## Code
 
-Each table has a repository type in `internal/database`:
+Each repository type in `internal/database` serves one or more tables:
 
-- `FileRepository`
+- `FileRepository`: `files`
-- `ChunkRepository`
+- `ChunkRepository`: `chunks`
-- `FileChunkRepository`
+- `FileChunkRepository`: `file_chunks`
-- `ChunkFileRepository`
+- `ChunkFileRepository`: `chunk_files`
-- `BlobRepository`
+- `BlobRepository`: `blobs`
-- `BlobChunkRepository`
+- `BlobChunkRepository`: `blob_chunks`
-- `SnapshotRepository`
+- `SnapshotRepository`:
-- `UploadRepository`
+    - `snapshots`
-- `LocalMetaRepository`
+    - `snapshot_files`
+    - `snapshot_blobs`
+    - `snapshot_mount_points`
+- `UploadRepository`: `uploads`
+- `LocalMetaRepository`: `local_meta`
 
-`Repositories` holds one of each.
+No repository type serves `schema_migrations`. The code that applies the
+migrations reads and writes it.
+
+`Repositories` holds one of each type.
 
 `Repositories.WithTx` runs a function in one transaction:
 
@@ -401,5 +414,5 @@ rolled back.
   `snapshot list`, can read the index while a backup writes to it.
 - A connection waits up to 10 seconds for a lock before it fails.
 - Foreign keys are enforced.
-- A process opens one connection, so all its access to the index is in turn.
+- Each open database uses one connection, so its queries run one at a time.
 - Queries pass their values as parameters.

docs/REPOSTRUCTURE.md

@@ -253,17 +253,23 @@ therefore does not stop overwrites.
 
 ### Sizes are visible
 
-age adds no padding. The size of an object is the exact size of its compressed
+An object's size follows exactly from the size of its compressed contents.
-contents.
 
 All the new chunks in one blob share one zstd stream. The stream looks back 8
 MiB, or 4 MiB at compression levels 1 and 2. A blob is closed at
 `blob_size_limit` and at the end of each configured path. A chunk that is
 already stored is never packed again.
 
-Someone who can write into a backed-up file and watch blob sizes can learn
+Data that an outsider writes compresses smaller when it matches a secret in the
-something in one case. Their data and a secret must land in the same chunk, in
+same chunk, so the blob gets smaller. The blob's size therefore shows the
-a file that keeps changing.
+outsider whether their data matched the secret.
+
+This works only when all of these hold:
+
+- The outsider can write into a backed-up file.
+- The outsider can see the blob sizes at the destination.
+- The outsider's data and the secret land in the same chunk.
+- The file keeps changing.
 
 Back up a directory that outsiders can write to as a snapshot of its own.
 
@@ -363,10 +369,11 @@ objects. An rclone remote that cannot move objects can keep a partial object.
 - a blob that no manifest lists
 - a directory under `metadata/` that has no manifest
 
+It reads the manifests, which are not encrypted, so it needs no private key.
+
 These rules keep it from deleting data that is in use:
 
 - It keeps every blob that any manifest lists.
-- It reads the manifests, which are not encrypted, so it needs no private key.
 - It deletes nothing when any manifest cannot be downloaded or decoded.
 - It deletes a blob only once the blob is older than `prune_grace_period`.
 - It deletes a directory only once every object in it is that old.