fix(cli): stop large sql blocks stalling db reset (CLI-2469)

[7ttp]

TL;DR

fixes db reset stalling on a seed with a large DO $$ block or /* */ comment.

prob

the statement splitter built each statement one character at a time, and on every character inside a $$ body or block comment it sliced the tail of that string.

in bun each slice copies the whole statement so far
so splitting grew quadratically and a 1 MB DO $$ block took about 15 seconds. seeds, migrations and db push all go through this splitter...

sol

tracking start and end offsets and slicing the original input instead,
so the split stays linear and the same 1 MB block takes about 40 ms. the split output is unchanged....

ref:

• closes: Slow seeding when having a large seed file and an anonymous block. #6686
• introduced in: feat(cli): port db push, db reset, and db start to native TypeScript #5715

[github-actions]

🤖 AI Review

The offset-based rewrite preserves splitter behavior and removes the identified quadratic accumulation path. Both Claude findings are confirmed as nits: the performance test relies on Vitest's implicit timeout, and the non-obvious offset/performance invariant is undocumented. Codex reported no findings, and no correctness or security defects were found.

Findings

Severity	Location	Category	Sources	Claim
⚪ NIT	apps/cli/src/command-internal/sql-split.unit.test.ts:164	test-coverage	claude	The performance regression test relies on Vitest's implicit default timeout instead of declaring its intended performance bound.
⚪ NIT	apps/cli/src/command-internal/sql-split.ts:242	maintainability	claude	The code does not document that slicing by start/end offsets avoids quadratic accumulation and that state-held offsets remain relative to the current token.

Stats

Claude findings: 2 · Codex findings: 0 · Confirmed: 2 · Refuted: 0 · Uncertain: 0

---

Models: claude-opus-5 + gpt-5.6-sol · Trigger: auto · Workflow run

This review runs once per PR. A maintainer can request another with a /ai-review comment.

[jgoux]

Cross-review with Claude Fable reached consensus after two rounds: no actionable findings introduced by this PR. Token-relative offsets, Unicode iteration, statement boundaries, and EOF behavior are preserved. No inline findings remain.

Validation: all 59 splitter assertions passed in an isolated Bun harness (test-runner import adapted from Vitest to bun:test); 20,000 generated SQL/Unicode inputs matched the base across all four exports; direct Bun measurements confirmed the large-block speedup. The normal local package test command was blocked by dependency setup. The PR CI checks are passing. The added tests verify large-input correctness; a durable automated performance bound was not established.