[thekevinbot]

Refs #1484

Sets PRAGMA threads = 8 on the ephemeral database so SQLite's sorter (GROUP BY / ORDER BY over a large table) runs on background threads.

Measurements (load 11-16, interleaved runs, 1M-row jsonl case)

config	min	median	peak RSS
base	0.642s	0.711s	~20MB
threads=8	0.579s	0.645s	~28MB
cache_size=64MB	0.604s	0.661s	~58MB
threads=8 + cache_size=64MB	0.601s	0.674s	~57MB

Decisions:

• Kept threads=8 only: about -9%, small RSS cost.
• Dropped cache_size and temp_store=MEMORY: no win beyond threads, 2-4x the RSS.
• Frontmatter (65k rows, tiny sort) and a plain path-table query: no change within noise (path query 21.1ms vs 21.6ms median).
• The on-file table is not an inserted table dirsql shapes for the user's GROUP BY, so no insert-order or layout lever was found; SQLite has no hash aggregate to ask for.
• Speed: jsonl parse is 1.4x the fastest native (jq per file under xargs -P) #1484 stays at about 1.34x the fastest native (jq+xargs-P) in the harness, so its bar is not flipped to FASTEST here.

Changelog / Migrations

• Changelog fragment added under packages/rust/changelog.d/
• Migration fragment with the same slug added under packages/rust/migrations.d/

E2E Verification

• Ran e2e suites locally for every affected SDK
• Python SDK e2e: pass
• TypeScript SDK e2e: pass
• Rust core e2e: covered by the two above
• packages/python/e2e-attestations/bot-speed-parse-pragmas-1011.json receipt written
• packages/ts/e2e-attestations/bot-speed-parse-pragmas-1011.json receipt written
• Command(s) run: pytest tests/e2e/{jsonl_parse_speed,frontmatter_parse_speed,on_file_query,query_contract}_test.py -x -q (14 passed); pnpm test:e2e (18 passed)
• Result summary: pass

Red run: 38187945310 (Unit tests failed on new_lets_the_sorter_use_worker_threads, the pragma reading its default).