perf: cache common h2 header name buffers

[mcollina]

Summary

Optimize the HTTP/2 response header path in client-h2 by:

• avoiding the object rest/spread copy used to strip :status
• skipping pseudo-headers inline during H2 header parsing
• caching Buffer instances for a bounded set of common response header names

The header-name cache is intentionally bounded to avoid unbounded memory growth from attacker-controlled header names.

Validation

• npx eslint lib/dispatcher/client-h2.js
• npm run test:h2:core

Benchmark notes

I verified this with a controlled A/B .request() harness against main using the same simple H2 server and interleaved runs.

Average over 5 runs (80 rounds x 200 parallel requests):

• this branch: 2401.39ms
• main: 2452.65ms

That is roughly a 2.1% improvement on this request-heavy H2 workload.

[trivikr]

There's an alternative PR which avoids response headers reserialization in #5085, which shows 5-7% improvement for similar tests.