fix(inbox-worker): don't Nak on a CountDocuments failure after the status write commits

Per PR #363 review: in UpdateUserStatus the guarded UpdateOne already commits the
status write before the MatchedCount==0 branch runs. That branch's CountDocuments is
only a logging aid (absent-account vs stale-event), so returning its error caused a
pointless Nak that re-ran the already-applied, idempotent write. Downgrade the
CountDocuments failure to a warn and continue (return nil) instead.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012X9qhQT4NwmCHjdndwNtFD

Author: claude

inbox-worker/main.go

@@ -186,14 +186,14 @@ func (s *mongoInboxStore) UpdateUserStatus(ctx context.Context, account, statusT
 		return fmt.Errorf("update user status for %q: %w", account, err)
 	}
 	if res.MatchedCount == 0 {
-		// No match means either the account isn't on this site (expected no-op for the
-		// all-sites fan) or a stale/duplicate event lost the guard. Distinguish so a
-		// genuinely-missing account stays visible instead of being silently swallowed.
+		// The UpdateOne above already committed; MatchedCount==0 is a correct no-op (account not on
+		// this site, or a stale event the guard rejected). CountDocuments only picks the warn
+		// message, so a failure here must not Nak/retry an already-applied write — log and move on.
 		count, cerr := s.userCol.CountDocuments(ctx, bson.M{"account": account})
-		if cerr != nil {
-			return fmt.Errorf("check user existence for %q: %w", account, cerr)
-		}
-		if count == 0 {
+		switch {
+		case cerr != nil:
+			slog.WarnContext(ctx, "user existence check failed, skipping unknown-account log", "account", account, "error", cerr)
+		case count == 0:
 			slog.WarnContext(ctx, "user_status_updated for unknown account, skipping", "account", account)
 		}
 	}