perf(test): clone a migrated template DB per TestD1Database instead of replaying 178 migrations

[JSONbored]

Summary

Every TestD1Database() construction executed the full migrations/*.sql chain (178 files, growing with every merge) into a fresh :memory: database — measured at ~640ms per construction, across ~1,500 construction sites: ~950s of pure schema replay per full run, and the reason ordinary tests sat in the 3s+ range (observed live on PR #8524's validate-tests).

This replaces the per-construction replay with a content-hash-keyed, fully-migrated template database file built once per migration-chain content and shared across all vitest workers; each TestD1Database gets its own copyFileSync clone (~1.5ms including a verified write — ~440× faster per construction).

Measured on the full unsharded suite (same machine, same tree):

	before	after
wall time	510s	171s (3.0×)
aggregate test time	4,288s	870s (4.9×)
heaviest file (queue.test.ts)	107s	17s

Design constraints the shape answers (each verified the hard way, documented in the helper):

• serialize()/deserialize() don't exist on pinned Node 22 (the prior attempt crashed CI) — file copy is Node-22-safe.
• Clones must NOT be unlinked while open: SQLite fails later writes with SQLITE_READONLY_DBMOVED. Clones stay on disk until a single per-process exit sweep.
• Vitest's per-file module isolation re-evaluates the helper constantly, so all memoization lives on globalThis — module-scope memos would rebuild the template once per test file and silently give back most of the win.
• Template built at a .tmp sibling then atomically renameSyncd — concurrent workers double-build harmlessly; no reader ever copies a half-written template; a schema change gets a fresh hash key.
• page_size=1024 + VACUUM shrink the empty-schema template 1.4MB → ~410KB, bounding worst-case tmpdir usage for a run's clones to a few hundred MB, swept at exit.

Validation

• npm run test:coverage (full, unsharded): 21,472 passed / 0 failed — identical pass set to the pre-change baseline run, at 171s vs 510s.
• Only test/helpers/d1.ts changes (Codecov-ignored path); no runtime code touched.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

❌ 3 Tests Failed:

Tests completed	Failed	Passed	Skipped
21448	3	21445	21

View the top 3 failed test(s) by shortest run time

test/unit/rag-index.test.ts > indexRepo: MAX_CHUNKS_PER_REPO cap holds > still indexes package.json (and other root manifest/config files) on a repo whose file count exceeds the cap (regression: manifestPriority)

Stack Traces | 36s run time

Error: Test timed out in 15000ms.
If this is a long-running test, pass a timeout value as the last argument or configure it globally with "testTimeout".
 ❯ test/unit/rag-index.test.ts:530:3

test/unit/rag-index.test.ts > indexRepo: MAX_CHUNKS_PER_REPO cap holds > stops upserting once the per-repo cap is reached

Stack Traces | 37.2s run time

Error: Test timed out in 15000ms.
If this is a long-running test, pass a timeout value as the last argument or configure it globally with "testTimeout".
 ❯ test/unit/rag-index.test.ts:511:3

test/unit/rag-index.test.ts > reindexChangedPaths: delete + re-upsert only the changed paths > caps incremental reindex upserts at MAX_CHUNKS_PER_REPO

Stack Traces | 43.8s run time

Error: Test timed out in 15000ms.
If this is a long-running test, pass a timeout value as the last argument or configure it globally with "testTimeout".
 ❯ test/unit/rag-index.test.ts:735:3

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}