Bound unique constraint row checks

[joe-thom]

What changed

Unique-constraint checks now stop after the first matching row instead of counting every row. This reduces work on PostgreSQL read replicas and preserves the empty-table decision.

Why

The Replit compatibility adapter runs this query before it asks how to handle a new unique constraint. Full-table counts increase schema-check latency and the chance of a PostgreSQL recovery conflict.

References CLO2-222
Project: DB Sprint

Validation

• pnpm --dir drizzle-kit exec vitest run tests/api.test.ts
• pnpm --filter @drizzle-team/drizzle-kit test:types
• pnpm lint
• pnpm format:check
• pnpm --filter @drizzle-team/drizzle-kit build

Rollback

A revert is safe. This change does not write database state.

Stack

• Bring drizzle-kit fork to v1 RC #26 adds the Replit compatibility API for drizzle-kit v1 (1/2)
• Bound unique constraint row checks #30 bounds its unique-constraint row check (2/2), this PR

~ written by Zerg 👾 (wp-cc5da088)
Linear project: DB Sprint