fix(backend): report ingest correctness — propagate storage failure, 413 on oversize, global body cap (closes #23)

A buffer failure on POST /api/v1/reports now returns 500 instead of a
false {"status":"ok"}, so clients can retry. Decode errors are split:
an over-limit body returns 413 (via errors.As on *http.MaxBytesError),
malformed JSON stays 400. A new MaxBodyBytes middleware caps every
route (1 MiB default; rejects an oversized Content-Length up-front and
caps the read otherwise), replacing the per-route reader so the health
check and future routes are bounded too. The raw attacker-controlled
geo blob is no longer logged — only its byte length — and client_id and
timestamp are length-bounded before logging. A decodeJSON handler helper
is added per the HTTP server conventions. Panic recovery is now a local
middleware that routes the stack through slog as structured JSON rather
than chi's plain-text stderr. Error bodies still leak nothing internal.

Chosen failure code for a storage failure: 500, since a full buffer or
write error is server-side and retryable, not the client's fault.

Model: opus-4-8

backend/internal/middleware/export_test.go

@@ -1,6 +1,7 @@
 package middleware
 
 import (
+	"log/slog"
 	"net/http"
 	"net/netip"
 )
@@ -8,6 +9,12 @@ import (
 // Test-only wrappers exposing unexported helpers to the
 // external middleware_test package.
 
+// NewWithLogger builds a Middleware around a logger for tests
+// that exercise the logging paths without the fx graph.
+func NewWithLogger(log *slog.Logger) *Middleware {
+	return &Middleware{log: log}
+}
+
 func ClientIP(
 	remoteAddr string,
 	header http.Header,