fix(lcm): structural FTS rebuilds log why + how long; undecryptable-row fixture (follow-up to #966)

[Kyzcreig]

Follow-up to #966 (card t_d3963974, Apollo boot-cost freeze #3).

Why

The two structural FTS rebuilds on 2026-09-24 (05:07→05:20 = 13 min, 05:22→05:28 = 6.5 min, each under _LOAD_LOCK + the SQLite write lock, up to 12 turns queued) left no log line at all — the only trace was PHASE=context_engine_load_slow minutes later, and py-spy. repair_external_content_fts decided to DROP + INSERT ... VALUES('rebuild') silently.

What

• repair_external_content_fts: when a structural rebuild is about to run, log why (missing table / missing shadow <name> / not fts5 / column missing / schema-probe error: <exc> — the last one covers a transient SQLITE_BUSY on the sqlite_master probe, which the old code also answered with a rebuild) plus an O(1) size hint (max(rowid) of the content table), then log the duration after the rebuild commits. _fts_needs_rebuild_structural now records its reason in _last_structural_reason for the caller.
• Test fixture (test_lcm_init_cost_regression.py::_fill): one AEAD-prefixed row the (disabled) test cipher cannot decrypt — the fleet DB has 3 such rows — so test_engine_construction_needs_no_write_lock now also gates the per-boot NULL-over-NULL msg_fts_update trigger write that fix(lcm): FTS parity COUNT(*) ran under _LOAD_LOCK on every engine load (freeze #3) #966 removed (4–57 s per boot on the fleet DB, 3 row(s) undecryptable × 813 in the gateway log).

Evidence

• tests/context_engine/: 395 passed on this branch (/tmp/lcm3-suite2.txt).
• Fixture RED on pre-fix(lcm): FTS parity COUNT(*) ran under _LOAD_LOCK on every engine load (freeze #3) #966 code: test_engine_construction_needs_no_write_lock fails at store.py:_backfill_search_content (database is locked); GREEN here.

No behavior change to the rebuild decision itself (that is #966); this is observability + a test gap.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[Kyzcreig]

🤖 merged-by: apollo · lane: boil-ocean-aged · gate: BYPASS: FR PAUSED by Ace ruling 2026-09-22 (state/fleetreview-pause-20260922.md); Apollo-reviewed lands via bypass · why: 0/1 commits on main by patch-id; content not in #1023 (no _fts_structural_problem on main); rebased clean onto 7ed45d9, re-authored to Kyzcreig noreply for check-attribution; 19 passed lcm fts/init_cost tests