fix: preserve indexed snapshots in dashboard history

[floze-the-genius]

Fixes #1307.

Summary

• use live indexed snapshots as fallback evidence for dashboard history when the matching backup operation is no longer present
• deduplicate fallback snapshots against backup operations by snapshot ID
• exclude forgotten snapshots while preserving per-day byte details and overdue tracking

Tests

• go test ./internal/api -count=1
• go test ./internal/orchestrator/tasks -count=1
• go vet ./internal/api
• go test ./... -count=1 (with the CI-style fake webui/dist/index.html.gz embed)

[CLAassistant]

All committers have signed the CLA.

[garethgeorge]

Hi, thanks for the interest in contributing -- would you mind including some screenshots / explaining a in a bit more detail the problem you're seeing and how this CL is addressing it?

[garethgeorge]

Okay looking deeper -- seeing a few lingering issues:

• The history strip gets fixed but the counters next to it don't -- the fallback never updates backupsSuccess30, backupsExamined, the byte totals, protectedBytes, or backupChart. Those all render on the same card (SummaryDashboard.tsx:596-623, and :179-190 where the status dot is derived from recentBackups), so in the case this is meant to fix you'd see green days sitting above "0 ok" with an idle dot.
• Both new maps hold every snapshot in the oplog, and twice over since each one goes into the plan acc -- this is quite a bit of memory and needs to be avoided. The start time of a snapshot will always be after the time the backup operation ran, and will be somewhat close, we should be able to do something more clever here.
• nit: applyIndexedSnapshotFallbacks mutates state from inside finalize, so calling finalize twice would double count. Probably cleaner to call it once at the end of the walk.

[floze-the-genius]

Updated the implementation in f7b7704 to cover the remaining dashboard state:

• indexed-snapshot fallbacks now update success/examined counts, byte totals and averages, protectedBytes, daily history, staleness evidence, and backupChart through the same accounting path;
• the newest-first oplog walk retains at most one unmatched indexed snapshot per plan instead of maps containing every snapshot;
• pending fallbacks are emitted in newest-first order;
• fallback application now runs once after the oplog walk, outside finalize, so repeated finalization does not double-count.

Added coverage for counters, bytes, chart entries, deduplication, forgotten snapshots, ordering, and repeated finalization. Validation: go test ./internal/api -count=1, go test ./internal/orchestrator/tasks -count=1, and go vet ./internal/api.

[floze-the-genius]

Controlled local before/after reproduction, using the same config and operation log for both binaries. The fixture contains three successful OperationIndexSnapshot records (4, 2, and 0 days ago) and no OperationBackup records.

Base 148ba12c: the dashboard ignores those live indexed snapshots and renders the empty state.

PR head f7b7704a: the same data renders Everything is backed up, three successful history days, Protected 300 MiB, and Last upload 30 MiB.

This verifies that the fallback now feeds the complete dashboard accounting path rather than only coloring the history strip. Validation on the PR head: go test ./internal/api and pnpm run build.

[floze-the-genius]

Hi @garethgeorge — following up on the requested changes: f7b7704 addresses the dashboard counters, chart, and staleness path; bounds indexed-snapshot retention; and moves fallback application out of finalize. The controlled before/after reproduction is above, and the branch is still current with main. Is there anything else you would like changed?