Implement internal/rollback and the service control it needs

Rollback was the one phase gating everything else: `run` without --dry-run
refused because this tool could not undo a cutover it had committed to.
That reason is now gone, and the refusal has narrowed to the fact that
there is no real cutover to undo yet.

internal/rollback implements ARCHITECTURE.md 4.8 as eight checkpointed
steps under PhaseRollback: verify-backup, stop-service,
preserve-failed-state, restore-data, restore-binary,
restore-service-config, start-service, verify-rollback.

Three things depart from what 4.8 specified, each for a reason:

- The backup is re-verified against its manifest *before* the service is
  stopped, which the design didn't call out. Finding a corrupt backup is
  survivable while the failed instance is still up, and unsurvivable once
  its data directory has been moved aside.
- BuildPlan is separate from Run, so every reason to refuse (closed
  rollback window, FoundationDB, no recorded backup, unknown deployment
  kind, missing database credentials) is found before anything is touched.
  The CLI prints that resolved plan and acts only with --yes.
- The restore is re-verified against the same manifest after writing. A
  restore that put back truncated bytes and reported success would be
  worse than one that failed outright.

Nothing from the failed attempt is deleted: the half-migrated data
directory and the displaced binary are moved to .failed-<run-id> names, so
a retry after the underlying issue is fixed still has both the evidence and
the artifacts. Afterwards a reduced validation suite runs against the
*restored* instance (version, reachability, directory counts) rather than
assuming the restore worked.

internal/service is a new package holding the systemd/Docker control this
needs. It's separate rather than living inside internal/rollback because
cutover will need the identical operations, and because the commands that
can take mail delivery down belong in one auditable place - the same
reasoning that makes stalwartapi the only thing speaking JMAP.
preflight.DeploymentKind is now a type alias for service.Kind so detection
and control can't drift apart. Its Active() reads `systemctl is-active`'s
output rather than its exit status: systemctl exits non-zero for every
non-active state, so exit-status logic would make "inactive" - the answer a
rollback most needs - look like a failure to read the state at all.

Also fixes a pre-existing bug in `status`: Go's flag package stops parsing
at the first positional argument, so `status <run-id> --state-dir X` looked
the run up in the default directory and reported it missing. `rollback`
would have inherited the same footgun on a command whose flags decide what
gets overwritten.

Still open: `confirm` cannot set RollbackWindowClosed. Rollback honours the
flag and refuses when it's set, but closing the window is the point of no
return for the backups this restores from, so it should land with the
retention policy 6 describes rather than before it.

Verified end to end against a fake systemd deployment: half-migrated data
restored to its original contents, failed state preserved, old binary
reinstalled and reporting 0.15.5, unit restarted, and a re-run of the
completed rollback inert.

README.md

@@ -8,9 +8,10 @@ Go, standard library only — no external dependencies.
 
 ## Status
 
-Partially implemented. Roughly 6,000 lines of tested code across backup,
-preflight, checkpointing, validation and recovery; two packages are still
-stubs, and that gap is what gates the rest.
+Partially implemented. Roughly 8,400 lines of tested code across backup,
+preflight, checkpointing, validation, recovery and rollback. What's missing
+now is the real cutover phase — the step that actually swaps the binary and
+switches the live service over.
 
 | Command | State |
 |---|---|
@@ -18,29 +19,35 @@ stubs, and that gap is what gates the rest.
 | `stalwart-migrate run --dry-run` | **Works** — preflight, real backup, sandboxed trial conversion |
 | `stalwart-migrate run` | **Refuses on purpose** — see below |
 | `stalwart-migrate status <id>` | **Works** |
-| `stalwart-migrate rollback <id>` | Not implemented |
+| `stalwart-migrate rollback <id>` | **Works** — prints its plan; acts only with `--yes` |
 | `stalwart-migrate confirm <id>` | Not implemented |
 | `stalwart-migrate report <id>` | Not implemented |
 
-**`run` without `--dry-run` deliberately refuses to proceed.** A real cutover
-needs `internal/rollback` — currently a `doc.go` and nothing else — plus real
-systemd/Docker service control. Committing to a migration with no working
-rollback would break the single guarantee the tool exists to make, so it
-stops rather than going partway. That refusal is the correct behaviour today,
-not a bug.
+**`run` without `--dry-run` deliberately refuses to proceed.** Rollback and
+service control now exist, so the reason has narrowed: what's still missing
+is cutover itself (ARCHITECTURE.md §4.5) — installing the new binary,
+rewriting the service definition, and starting the migrated instance for
+real. `run` stops rather than going partway. That refusal is the correct
+behaviour today, not a bug.
+
+Rollback was built before cutover on purpose: this tool should never be able
+to commit to a change it can't undo. `stalwart-migrate rollback <run-id>`
+resolves what it would do from the run's own checkpoint, prints that plan,
+and touches nothing without `--yes`.
 
 Package state:
 
 | Package | Lines | Tests |
 |---|---|---|
+| `internal/rollback` | 1807 | yes |
 | `internal/backup` | 1805 | yes |
-| `internal/preflight` | 1324 | yes |
+| `internal/preflight` | 1329 | yes |
 | `internal/validate` | 792 | yes |
 | `internal/stalwartapi` | 716 | yes |
 | `internal/recovery` | 702 | yes |
 | `internal/checkpoint` | 559 | yes |
+| `internal/service` | 467 | yes |
 | `internal/plan` | 196 | yes |
-| `internal/rollback` | stub | — |
 | `internal/config` | stub | — |
 
 ## Why not a shell script
@@ -92,3 +99,39 @@ directory — read the caveat the command prints before using it on anything
 you care about. Where the plan crosses the 0.15/0.16 boundary it clones that
 verified backup into a disposable sandbox and converts the copy, leaving the
 original untouched.
+
+## Rolling back
+
+`rollback` is the one command that stops a running mail server and
+overwrites a live data directory, so it never acts on its own reading of the
+situation without showing you that reading first:
+
+```sh
+stalwart-migrate rollback <run-id>          # prints the plan, touches nothing
+stalwart-migrate rollback <run-id> --yes    # performs it
+```
+
+Everything in the plan — which backup, which manifest, which preserved
+binary — comes from that run's own checkpoint, so rolling back days later
+doesn't depend on remembering any of it. What the checkpoint deliberately
+doesn't store (database credentials for an external SQL backend) is what the
+flags are for.
+
+The order matters and is not negotiable: the backup is re-verified against
+its manifest **before** the service is stopped, because a corrupt backup is
+survivable while the failed instance is still up and unsurvivable once its
+data directory has been moved aside. Nothing from the failed attempt is
+deleted — the half-migrated data directory and the displaced binary are
+moved to `.failed-<run-id>` names. Afterwards a reduced validation suite runs
+against the *restored* instance (version, reachability, directory counts)
+rather than assuming the restore worked; pass `--admin-url` to get the last
+two, which are skipped without it.
+
+Every step is checkpointed, so a rollback interrupted partway — which is
+exactly when a machine is most likely to be rebooted out from under it —
+resumes where it stopped instead of restarting a destructive sequence from
+the top. Re-running a completed rollback is inert.
+
+FoundationDB installs are refused rather than attempted: the backup phase
+only *starts* an `fdbbackup` job, and restoring one needs `fdbrestore`
+against a quiesced cluster.