Add benchmark suite comparing lambda-api to other Lambda frameworks (#34)

[naorpeled]

Closes #34.

Adds a benchmark suite comparing lambda-api against other popular AWS Lambda web frameworks.

Approach: in-process micro-benchmarks (not LocalStack)

The signal we care about is framework overhead — parse event → route → middleware → serialize — which the issue clocks at ~0.68 ms. This suite invokes each framework's compiled aws-lambda handler in-process, in the same Node VM, with identical synthetic API Gateway events, and measures it with mitata.

LocalStack / a real Lambda deploy was deliberately avoided for the primary suite: Docker + runtime bootstrap + network add tens of ms of noise (10–100× the signal) and would measure the emulator, not the framework. End-to-end deployment timing is a different question and is documented as an appendix in benchmarks/README.md.

What's included

• benchmarks/ — an isolated package with its own package.json, so the root library keeps its zero-dependency policy. It's excluded from the npm tarball by the existing files whitelist (verified with npm pack --dry-run).
  • run.js — the harness (mitata; a correctness gate validates each cell's status/body before timing, so no framework is ever benchmarked while silently 404-ing or throwing)
  • frameworks/ — baseline (raw handler), lambda-api, serverless-express (Express), fastify, hono, behind a uniform { name, version, build } contract
  • lib/ — v1/v2 event builders (seeded from the repo's __tests__/sample-event-* shapes), scenarios, validator, markdown renderer, README writer
• Scenarios: get-json, path-param, post-json, routing-50 (routing cost), not-found — each across API Gateway REST (v1) and HTTP API (v2).
• .github/workflows/benchmark.yml — runs on release published and manual dispatch (Node 20); refreshes the README Benchmarks section, Prettier-formats it, commits back to the default branch with [skip ci], and uploads results as an artifact. Not wired into PR CI, since perf is noisy on shared runners.

README updates on every release

The README gains a marker-delimited (<!-- BENCHMARKS:START/END -->) Benchmarks section that's regenerated in place (idempotent) with a compact one-row-per-release history table appended. Move the marker block anywhere once; future runs respect its position.

How to run locally

cd benchmarks
npm install
npm run bench

Notes for review

• The committed numbers are an example run on my machine (Apple M4 Max, Node 20) at version 0.0.0-development; the workflow refreshes both the numbers and the version (from the release tag) on each release.
• Added a standard catch-all 404 handler to the Express setup — without it, unmatched routes hit Express's finalhandler, which on-finished can't attach to serverless-express's mock socket (it 500s). This mirrors the built-in 404 of the other frameworks.
• benchmarks/ is added to .eslintignore/.prettierignore so the suite's dynamic import() (ESM-only hono/mitata) doesn't trip the root's ES2018 lint config.

Existing test suite (480 tests) and tsd types pass; root ESLint/Prettier are clean.