Backfill BM25/vector for pre-v3.1 events (v3.2 / 61-01)

[RichardHightower]

Tracking F5 / backfill from docs/plans/v3.2-prove-it-plan.md.

README BM25 row: events indexed before v3.1 have empty text_preview and there is no backfill command. admin rebuild-bm25 is a prune, not a rebuild (CHANGELOG / docs/UPGRADING.md).

An import path is a backfill path. March origin/gsd/phase-56-import-bootstrap already has streaming ExportBackup / ImportBackup plus memory backup / memory import. Cherry-pick by feature after 59-02 (see docs/plans/phase-59-orphan-branch-triage.md); do not merge the branch wholesale (it would regress memory-orchestrator and memory-bench).

Work (61-01)

• memory-daemon admin backfill-index --index bm25|vector|all: replay events from RocksDB into the index, reset the index checkpoint, resumable, idempotent, progress on stderr.
• Build on the import/bootstrap code if the 59-02 cherry-pick landed it.
• Update the README status row and docs/UPGRADING.md to say backfill exists.

Success

An e2e test ingests events with an old daemon build's schema (fixture), runs backfill, and memory search returns previews. A pre-v3.1 store can be backfilled without data loss.

[RichardHightower]

61-01 is up as PR #51 (stacked on #49).

memory-daemon stop
memory-daemon admin backfill-index --index bm25 --from-sequence 0

--from-sequence 0 is the empty-preview path (outbox already drained). Default start resumes from the checkpoint; a second run reports 0 new. --dry-run writes nothing. The command refuses if the daemon PID is live.

Not rebuild-indexes (TOC/grips) and not rebuild-bm25 (a prune). e2e: 50 events, no indexer, backfill, TeleportSearch returns non-empty previews.

Import.rs was not cherry-picked — it is not on this line. Event-log iteration is the replay.