…rics; completion write persists the settled tier (#76, #77)

Ships in the not-yet-archived 1.8.0 (build 33 unchanged, never uploaded).

#76: BackgroundSyncService.completeSync persisted "idle"/"error" directly,
bypassing the issue-derived tier from #73 - a successful background run
(scheduled refresh, overnight catch-up, relay wake-up) overwrote an honest
attention snapshot (parked share, disconnected required peer, collision)
with a green idle until the next foreground open. The static background
context cannot recompute unresolvedIssues (the bridge is already stopped
on the background-owned path; peer grace state lives only in the manager),
so the foreground now persists an issue floor into the app group on every
widget write: max severity of unresolvedIssues plus unreachable folders,
excluding the kinds a background run itself invalidates (.staleSync - a
successful sync resolves staleness by definition; .backgroundSync - the
completion write knows the fresher outcome), so background-only operation
cannot stick a false amber. completeSync maps its result to the same
severities backgroundSyncIssueItem surfaces in-app, folds the floor in,
and derives the tier through SyncHeaderModel.deriveWidgetStatus - one
cascade for every status surface (decisions 012/013). An unknown persisted
floor decodes as warning, never none. One deliberate visible change: a
failed background sync now shows amber "Needs Attention" (the tier the
header shows for it on next open) instead of red "Sync Error". Wire
compatibility unchanged: fromWire still maps anything unknown to
attention, so no combination decodes into a false green.

#76 follow-up: the same write also read the folder count and synced-file
events only after cleanupBackgroundManaged had stopped the bridge - which
then reports an empty folder list, so every background-owned completion
rendered "Vaults: 0" on the widget and lost the files count - and stamped
its own timestamp as "last sync" for failed runs too. The completion write
now runs before cleanup (the values are read while the bridge is still
up), a stopped bridge falls back to the previous snapshot's folder count
instead of zero, and only a successful run stamps the last-sync triple -
a failure carries the previous one forward, so "last synced" keeps naming
the last real sync. Assembly is a pure function (fallback matrix fully
unit-tested).

#77: updateWidgetSyncMetrics derived the completion tier while
activeWidgetSyncStart was still non-nil, so completeWidgetSyncSession
persisted a stale .syncing snapshot that the poll-end write immediately
replaced - two snapshot writes and widget reloads per sync completion.
The session is now finalized before the tier is derived; the poll-end
write finds an identical snapshot and dedupes to a no-op.

No strings added, no Go changes - the earlier-1.8.0 xcframework rebuild
note still applies before archiving.

Verified: full Xcode test plan green (iPhone 17 Pro simulator, 257/257
incl. the 20 new tests); the two #77 tests were demonstrated red on the
unfixed code first, the #76 defects are traced in the Go bridge (stopped
engine reports "[]" for folders and events).

Fixes #76
Fixes #77