v0.148.0 — coherent snapshot pairs + an offsite restore that actually restores (R-43 + R-44)

Closes the two findings from DIAG-immich-restore-2026-07-19. Viktor deleted 11
immich photos to test offsite restore; both runs flashed success and the photos
stayed gone. Two independent defects.

R-43 — no offsite path could restore a database. All three buttons were
file-only: the two "visszaállítás" actions staged to a scratch folder and never
touched postgres, and place-to-live merged only MISSING files. For a DB-indexed
app the bytes returned and the app still could not see them. The dump was
carried INTO every snapshot and could never be replayed OUT of one.

New ReconstituteFromOffsite (/backup/offbox/reconstitute): safety dump → stop →
files overwritten to the snapshot version → start → the snapshot's own dump
replayed → health wait. Two invariants:
  - nothing is ever deleted (-a, no --ignore-existing, no --delete): a file
    created after the snapshot survives as an extra;
  - the undo exists before the act — the pre-restore- dump is verified ON DISK
    before anything is stopped, overwritten or replayed; if it cannot be taken
    the operation refuses with zero changes.
The replay reads the SCRATCH unit: the live unit is never overwritten, so
replaying from it would replay the current DB over itself and restore nothing.

R-44 — a manual push shipped an unrefreshed dump (up to ~24h old). That day's
predated the customer's account by four hours and probed to asset:0/user:0/
album:0 inside 52MB whose bulk was immich's shipped geodata. Every run, manual
AND nightly, now refreshes dumps + units BEFORE capturing. Order is the
mechanism: the gap can only ADD files the DB does not reference yet, never
remove one it does. Manifests carry offsite_run_id + dumps_at, so coherence is
verifiable at restore time rather than assumed; the periodic refresh carries a
prior stamp forward and never invents one.

Honesty surfaces, all warn-level and none a gate: unstamped (pre-v0.148) pairs
report their skew, ValidateDump gained an EXACT-match accounts-table sniff for
customer-empty dumps, the completion flash states an outcome instead of a
mechanism, and the missing-only button now says what it does NOT do.

11 tests; 5 red-proofs run and reverted. Two of those found real test weaknesses
rather than confirming strength — the first undo mutation was caught by a second
guard, and the first table-matching test did not discriminate between the two
matchers at all. Both tests were rewritten to the cases that separate them.

NOT in scope: R-41's catalog invariant check, nightly cadence, retention, quota
math, tier-2, and v0.147.x progress semantics beyond one added phase line.

Live acceptance (§9) has NOT run: no capability-map flip, customer-restore row
stays MISSING, R-3 stays DRAFT.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P9Nn14TWGzKoqAJAiVwC2s

REUSE.md

@@ -205,6 +205,12 @@
 | `Manager.sambaUpFn` / `sambaPasswdFn` / `sambaRunFn` (func seams) | controller/internal/stacks/manager.go (fields) + samba.go | nil → `composeUp` / `docker exec smbpasswd` (STDIN) / `containerRunning("felhom-samba")` | injected in controller/internal/stacks/samba_test.go — the idempotency test asserts the up-seam is called **zero** times when config is unchanged; the passwd seam means no unit test ever handles a real secret or touches docker |
 | `Manager.sambaImgFn` (func seam) | controller/internal/stacks/manager.go (field) + samba.go | nil → `docker image inspect <infra.SambaImage>` | drives the 4b card's pulling-vs-starting decision, which MUST be taken before `compose up` (afterwards the image is always present) |
 | `Manager.offboxStreamRunner` + `SetOffboxStreamRunner` | controller/internal/backup/offbox_progress.go | nil → `defaultOffboxStreamRunner` (real `restic`, stdout scanned live) | streaming sibling of `offboxRunner`; fakes emit canned `--json` status lines in offbox_progress_test.go, so the whole progress path runs with no restic, network or repo |
+| `Manager.offsitePreDumpFn` + `SetOffsitePreDumpFn` (R-44, v0.148.0) | controller/internal/backup/offbox_reconstitute.go (seam) + offbox.go (call site) | nil → `runDBDumpsInternal` under the SAME running flag | THE dumps-before-capture ordering seam. Extracted so the order is observable without Docker/restic — an ordering guarantee no test can see is one refactor from silently reverting to the DIAG-immich-restore-2026-07-19 behaviour. Red-proof: moving the capture first yields `[capture dump]` |
+| `Manager.offboxFullPlaceCopier` + `SetOffboxFullPlaceCopier` (R-43) | controller/internal/backup/offbox_reconstitute.go | nil → `rsyncRestoreOverwrite` (`-a --itemize-changes`; **no** `--ignore-existing`, **no** `--delete`) | **TRAP: do NOT reuse `offboxPlaceCopier` here.** The two copiers have OPPOSITE semantics for an existing file — `--ignore-existing` is exactly what a full restore must not do, and conflating them is how a missing-only merge came to be labelled a restore. Never `rsyncMirror` (`--delete`) in any restore direction |
+| `Manager.safetyDumpFn` + `SetSafetyDumpFn` (R-43) | controller/internal/backup/offbox_reconstitute.go | nil → `DumpOne` | the pre-restore undo. Invariant: the `pre-restore-`-prefixed dump must be verified ON DISK before anything is stopped/overwritten/replayed; failure ⇒ refuse with zero changes. Red-proof requires removing BOTH guards (the `err != nil` return and the `os.Stat`) — removing one leaves the other holding |
+| `reimportDBDumpsFrom(ctx, stack, dumpDir)` | controller/internal/backup/restore_db.go | explicit-dir sibling of `reimportDBDumps` (which passes `AppDBDumpPath`) | offsite reconstitution replays from the SCRATCH unit: the live unit is deliberately never overwritten, so replaying from it would replay the current DB over itself and restore nothing |
+| `Manager.OffsiteScratchPair` / `OffsitePairInfo` | controller/internal/backup/offbox_reconstitute.go | reads the restored scratch unit's manifest (`offsite_run_id` / `dumps_at`) + the R-44 sniff | the confirm-dialog honesty surface. All warn-level: a pre-v0.148 (unstamped) pair and an empty-looking dump are SURFACED, never blocked — a false positive that refused a legitimate restore would be worse than the skew |
+| `appbackup.DumpValidation.LooksEmpty` (R-44 sniff) | controller/internal/appbackup/dbdump.go | computed in ValidateDump's existing single pass; `userTableNames` is EXACT-match | size and table count are both useless as emptiness heuristics (the 2026-07-19 dump: 52MB, 60+ tables, zero users — all geodata). **TRAP: never widen to a substring match on "user"** — it would flag `user_metadata` / `album_user` / `user_audit` on every healthy single-user box. A row wider than the read buffer still counts as a row |
 | `report.SetPendingControllerLog` / `SetControllerLogSource` | controller/internal/report/selftail.go | ACK-armed consume-once self-log pull (the logtail.go shape) | selftail_test.go; source = `logBuffer.Lines`, wired once in main.go |
 | `util.ParseVersion` / `util.Version.Compare` | controller/internal/util/version.go | THE one semver comparator (house rule: never a second) — selfupdate aliases it; agentapi's MinAgent comparison uses it | rejects pre-release/dev/latest (callers fall back, never trust); numeric compare (0.100 > 0.81) |
 | `agentapi.AgentVersionReporter` + `featureMinAgent` | controller/internal/agentapi/features.go | version-first Supports (v0.82.0 header channel); probe = fallback for header-less agents | a coupled feature adds BOTH a featureProbes row AND a featureMinAgent row; v0.116.0: `SupportsWithSource` also reports HOW the verdict was reached (version/probe-cache/probe) for the gate log line |