Verify login credentials before spending rate-limit budget (closes #150)

With TRUSTED_PROXIES empty behind the reverse proxy production is
required to run behind, every login POST keyed on the proxy's address
and shared one 5/minute bucket. A stranger sending five POSTs a
minute -- 0.08 requests per second, from anywhere -- kept that bucket
permanently full, and the operator's own correct password was answered
429 indefinitely with no second administrative path.

The login POST no longer has a pre-emptive limiter. The handler
verifies credentials first and spends budget only on a FAILED attempt,
so a correct password is never throttled whatever the counters hold.
Three things follow, and are implemented together because the first is
unsafe without the other two:

- Failures are counted per (client bucket, submitted username), five
  per minute, after which further failures get 429 with a Retry-After.
  A successful login clears the counter, so mistyping and then
  succeeding does not leave the operator throttled.
- Both key sets are capped at 1024 entries. The submitted username is
  attacker-controlled, so past the first cap failures fall back to a
  counter keyed on the client alone, and past both caps a failure is
  answered as throttled without being recorded. Tracked state stays
  under half a megabyte and does not grow with invented usernames.
- Concurrent Argon2id verifications are capped at two, a 128 MB
  ceiling at 64 MB per hash, and the queue for those slots is capped
  at 64 waiters. Every password-hashing endpoint takes a slot,
  including the password-change endpoint, which holds one across both
  its hashes. A request that waits five seconds without a slot is
  answered 503, and one that arrives with the queue already full is
  shed with 503 immediately rather than joining it. Bounding the wait
  alone would not bound memory: a waiter reaches the guard with its
  form parsed, so it holds up to the 1 MB body cap for the whole wait,
  and at flood rates an unbounded queue is worth gigabytes against a
  128 MB hashing budget. 64 waiters is 64 MB of committed queue
  memory, shallow enough that two slots drain a full queue inside the
  five-second deadline; peak commitment is 128 MB of hashing plus
  about 66 MB of parsed bodies.

An unknown username is verified against a dummy hash instead of
returning early, so a nonexistent account costs the same time as a
real one and the response cannot be used to enumerate usernames.

The password-change limiter is unchanged: RequireAuth runs ahead of
it, so only a request already carrying a valid session reaches its
bucket.

Two consequences are documented rather than fixed, because they follow
from the shape the issue asks for. Online guessing throughput rises
from 5 a minute to roughly 27 a second, about 2.3 million a day: the
credential check always precedes the counter, so the 429 is a label on
the response rather than a gate in front of the hash, and what bounds
brute force is the semaphore. And under a sustained flood the residual
exposure is a loss of login availability, not merely of latency --
above about 27 requests a second most attempts are shed with 503, so a
determined flood still denies login for as long as it runs. It costs
roughly 400x more to run, nothing accumulates, and the first attempt
after it stops succeeds. Restarting the service does not help: the
counters a restart clears are not what is saturated.

Also adds the missing test for the third bucketKey call site, where
the peer is a trusted proxy but the forwarded chain names no client.
Every existing test of that fallback uses an IPv4 proxy, where
bucketKey is the identity function, so dropping the /64 masking there
left the suite green.

README and the TRUSTED_PROXIES startup warning updated: a shared
bucket now costs precision, not the availability of the admin path.

internal/handlers/auth.go

@@ -2,6 +2,7 @@ package handlers
 
 import (
 	"net/http"
+	"strconv"
 
 	"sneak.berlin/go/webhooker/internal/database"
 )
@@ -93,6 +94,16 @@ func (h *Handlers) renderLoginError(
 
 // authenticateUser looks up and verifies a user's credentials.
 // On failure it writes an HTTP response and returns an error.
+//
+// The credential check runs BEFORE any rate-limit budget is
+// consulted, and only a failed check spends budget. That is what
+// keeps the single administrative path reachable: behind the reverse
+// proxy this deployment requires, with TRUSTED_PROXIES unset, every
+// client shares one bucket, so a limiter spent on arrival lets any
+// stranger deny the operator's own correct password indefinitely.
+//
+// Verifying first means every login POST costs an Argon2id hash, so
+// the work is taken under a bounded number of verification slots.
 func (h *Handlers) authenticateUser(
 	w http.ResponseWriter,
 	r *http.Request,
@@ -100,16 +111,37 @@ func (h *Handlers) authenticateUser(
 ) (database.User, error) {
 	var user database.User
 
+	release, ok := h.mw.BeginPasswordVerification(r.Context())
+	if !ok {
+		h.log.Warn(
+			"password verification capacity exhausted",
+			"path", r.URL.Path,
+		)
+		h.renderLoginError(
+			w, r,
+			"The server is busy verifying credentials. "+
+				"Please try again.",
+			http.StatusServiceUnavailable,
+		)
+
+		return user, errVerificationBusy
+	}
+
+	defer release()
+
 	err := h.db.DB().Where(
 		"username = ?", username,
 	).First(&user).Error
 	if err != nil {
+		// A username that does not exist is charged the same work
+		// as one that does. Skipping the hash here would answer in
+		// microseconds where a real account takes tens of
+		// milliseconds, handing every client a username oracle.
+		h.dummyVerifications.Add(1)
+		database.VerifyDummyPassword(password)
+
 		h.log.Debug("user not found", "username", username)
-		h.renderLoginError(
-			w, r,
-			"Invalid username or password",
-			http.StatusUnauthorized,
-		)
+		h.rejectLogin(w, r, username)
 
 		return user, err
 	}
@@ -127,16 +159,49 @@ func (h *Handlers) authenticateUser(
 
 	if !valid {
 		h.log.Debug("invalid password", "username", username)
+		h.rejectLogin(w, r, username)
+
+		return user, errInvalidPassword
+	}
+
+	// The password was correct, so forgive whatever failures this
+	// client accumulated: an operator who mistypes a few times and
+	// then gets it right must not stay throttled afterwards.
+	h.mw.ForgiveLoginFailures(r, username)
+
+	return user, nil
+}
+
+// rejectLogin counts one failed credential verification and answers
+// it: 401 while this client still has failure budget against the
+// submitted username, 429 with a Retry-After once it is spent.
+//
+// The 429 throttles wrong passwords only. A correct one never
+// reaches here, so no amount of failure — from this client or any
+// other sharing its bucket — can keep the operator out.
+func (h *Handlers) rejectLogin(
+	w http.ResponseWriter,
+	r *http.Request,
+	username string,
+) {
+	if !h.mw.RecordLoginFailure(r, username) {
 		h.renderLoginError(
 			w, r,
 			"Invalid username or password",
 			http.StatusUnauthorized,
 		)
 
-		return user, errInvalidPassword
+		return
 	}
 
-	return user, nil
+	w.Header().Set("Retry-After", strconv.Itoa(int(
+		h.mw.LoginFailureInterval().Seconds(),
+	)))
+	h.renderLoginError(
+		w, r,
+		"Too many failed login attempts. Please try again later.",
+		http.StatusTooManyRequests,
+	)
 }
 
 // createAuthenticatedSession regenerates the session and stores