fix(repair): preflight SQLite integrity before rebuild

[fatkobra]

What does this PR do?

Fixes #1362.

This makes mempalace repair --yes run a SQLite PRAGMA quick_check before it reaches Chroma's destructive delete_collection() rebuild step.

If SQLite reports corruption, repair now stops with a clear recovery message instead of surfacing an opaque Chroma/SQLite code 779 traceback from inside delete_collection().

What changed

• Added sqlite_integrity_errors().
  • Opens chroma.sqlite3 read-only.
  • Runs PRAGMA quick_check.
  • Returns any non-ok messages as repair-blocking errors.
• Added print_sqlite_integrity_abort().
  • Prints a clear explanation that SQLite-layer corruption was detected.
  • Points the user toward offline SQLite recovery before re-running repair.
• Wired the preflight into repair.rebuild_index() after extraction/truncation safety and before:
  • backing up chroma.sqlite3
  • backend.delete_collection(...)
  • collection recreation
• Added regression tests for:
  • healthy SQLite quick_check
  • unreadable SQLite file reporting
  • repair --yes aborts on SQLite-layer index corruption (code 779) before reaching HNSW rebuild #1362-style abort before delete_collection()

Why

#1362 reports a palace where HNSW was not the corrupt layer. The HNSW segment looked healthy, but repair --yes crashed inside Chroma's delete_collection() with:

(code: 779) database disk image is malformed

## How to test
ruff format mempalace/repair.py tests/test_repair.py
ruff check mempalace/repair.py tests/test_repair.py
python -m pytest tests/test_repair.py -q
python -m pytest tests/ -v

## Checklist
- [x] Tests pass (`python -m pytest tests/ -v`)
- [x] No hardcoded paths
- [x] Linter passes (`ruff check .`)

[igorls]

Maintainer-edit: merged develop in to resolve conflicts opened by #1285 (temp-staging rebuild) and #1312 (collection_name plumbing) after this branch was authored.

• `mempalace/repair.py`: kept this branch's `sqlite_integrity_errors()` and `print_sqlite_integrity_abort()`; took develop's `rebuild_index` signature with the `collection_name` parameter. Normalized the helper's print indent to 2 spaces to match the rest of the file.
• `tests/test_repair.py`: kept both this branch's preflight tests and develop's `rebuild_from_sqlite` / configured-collection coverage. Took develop's temp-staging assertion shape for `test_rebuild_index_success`.
• Test fixture update: 7 `test_rebuild_index_*` tests used `sqlite_path.write_text("fake")` — that fails PRAGMA quick_check, so the new preflight aborted before the rebuild logic the tests actually exercise. Replaced with `sqlite3.connect(str(sqlite_path)).close()` (empty real SQLite DB that passes quick_check), preserving the tests' intent.

Local: 1618 tests pass, ruff lint+format clean against the 0.4.x CI pin.

[igorls]

Round-2 merge: #1357 (max_seq_id preflight) merged into develop while this branch was in CI, opening a fresh conflict between the two preflight helpers. Kept both — they check distinct corruption classes:

• `sqlite_integrity_errors` (this PR) — PRAGMA quick_check before `delete_collection()`
• `maybe_repair_poisoned_max_seq_id_before_rebuild` (fix(repair): preflight poisoned max_seq_id before rebuild #1357) — un-poisons max_seq_id rows so queued writes aren't dropped

Both run as separate preflights inside `rebuild_index`; non-overlapping coverage in tests too. Local: 1623 tests pass on the merge state, ruff lint+format clean.