Multiple error paths in the sync pipeline silently swallowed errors,
allowing sync to report IN_SYNC while clients had diverged state.
This caused permanent data loss — tasks missing on one device but
present on another, with no indication anything was wrong.
Fixes:
1. Propagate download failure: when OperationLogDownloadService
returns success=false, throw instead of treating as "no new ops".
Prevents lastServerSeq from advancing past failed downloads.
2. Throw on LWW conflict apply failure: autoResolveConflictsLWW
now throws when applyOperations fails, matching the behavior of
applyNonConflictingOps. Prevents lastServerSeq from advancing
past failed conflict resolutions.
3. Propagate rejected ops handler errors: rethrow instead of
swallowing in the uploadPendingOps catch block, so the
sync-wrapper can set ERROR status.
4. Surface validation failure: validateAfterSync now checks the
return value from validateAndRepairCurrentState and shows an
error snackbar when validation fails.
5. Set ERROR status in sync-wrapper catch-all: the generic error
handler now calls setSyncStatus('ERROR') so the sync icon shows
the red error state.
Fixes#6571