feat: add test coverage reporting with vitest

[schlpbch]

Summary

Add test coverage reporting infrastructure to the conformance test suite using vitest's built-in coverage support.

• Install @vitest/coverage-v8 provider for code coverage analysis
• Configure vitest with coverage settings (text, JSON, HTML reporters)
• Add test:coverage script for convenient coverage runs
• Add coverage/ to .gitignore
• Current coverage: 59.16% statements, 45.87% branches

Test Plan

• Run pnpm test:coverage - all 84 tests pass with coverage report generated
• Verify coverage reports generated in text, JSON, and HTML formats
• Check that coverage excludes node_modules, dist, test files, and examples

🤖 Generated with Claude Code

[schlpbch]

@maciej-kisiel could you please review this PR?

[pcarleton]

Thanks for the contribution! I'm going to close this one though. I don't think code coverage is the right metric for this repo.

The conformance suite is a test harness: its job is to exercise SDK implementations against the MCP spec. The coverage number that matters to us is spec coverage (how much of the protocol the scenarios test), not line coverage of the harness itself. We have work to do to track that still.

Line coverage here tends to encourage low-signal tests: things like asserting scenario.name === 'initialize' or that a color-mapping function returns the right color for each enum value. Those pad the number but don't catch real bugs, and they add maintenance burden every time we tweak a string.

The integration tests we already have (all-scenarios.test.ts, checks.test.ts) exercise the scenarios end-to-end, which is the granularity that actually catches regressions in this kind of project.

(Also, small note: this repo uses npm rather than pnpm, so the pnpm-lock.yaml would conflict with the existing package-lock.json.)

Appreciate you taking the time to put this together — if you're interested in contributing, adding new conformance scenarios (especially for under-covered parts of the spec) would be really valuable.

[schlpbch]

Sure, scenarios would be fine as well.

I'd argue that also in the scenario space it would be interesting to have a metric in how far the specification is covered by scenarios.