Test matrix: integration tests, Playwright, benchmarks, coverage

[thedavidweng]

Problem

No integration tests for the generation state machine or model download flow. No E2E smoke tests. No coverage enforcement in CI. The test pyramid has a wide base (unit) but missing middle (integration) and top (E2E).

Scope

• Rust integration tests for model download and generation state machine
• Playwright smoke tests (3 happy paths: setup, generate, history)
• Performance benchmark script (scripts/bench.mjs) with markdown output for release notes
• Database migration tests (schema v1 → latest)
• Playwright screenshot comparison for key pages
• Expand frontend unit tests from 42 to 120+ across 16+ files
• Add Vitest coverage via @vitest/coverage-v8
• CI coverage reporting and threshold enforcement (60% minimum)

Acceptance criteria

• Generation state machine has integration tests covering happy path, failure, and cancellation
• Playwright runs 3 smoke tests in CI
• pnpm test:coverage enforces 60% line coverage minimum
• Migration tests verify schema upgrades from v1
• Benchmark script produces markdown output suitable for release notes