perf: lazily materialize raw h2 headers

[mcollina]

Summary

This stacks on top of #5122.

It keeps the bounded common-header buffer cache, and adds a follow-up optimization for the HTTP/2 response path:

• pass parsed response headers as an object to the H2 request handlers
• defer raw header materialization until controller.rawHeaders is actually accessed
• update the API handlers to only read controller.rawHeaders when responseHeaders === 'raw'

The goal is to avoid paying the parseH2Headers() allocation cost on the hot path when callers only need the normal parsed header object.

Validation

• npx eslint lib/core/request.js lib/dispatcher/client-h2.js lib/api/api-request.js lib/api/api-stream.js lib/api/api-connect.js lib/api/api-upgrade.js lib/api/api-pipeline.js
• npm run test:h2:core
• npm run test:h2:fetch

Benchmark notes

I verified this against commit 63a1b0ff from #5122 using the same controlled H2 .request() harness and interleaved runs.

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

• 63a1b0ff: 2409.08ms
• this branch: 2396.64ms

That is roughly a 0.5% improvement on top of the previous optimization.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 75.86207% with 14 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (perf/h2-header-buffer-cache@63a1b0f). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
lib/api/api-pipeline.js	25.00%	6 Missing ⚠️
lib/api/api-stream.js	25.00%	3 Missing ⚠️
lib/api/api-upgrade.js	25.00%	3 Missing ⚠️
lib/api/api-connect.js	33.33%	2 Missing ⚠️

Additional details and impacted files

@@                      Coverage Diff                       @@
##             perf/h2-header-buffer-cache    #5123   +/-   ##
==============================================================
  Coverage                               ?   93.11%           
==============================================================
  Files                                  ?      110           
  Lines                                  ?    35992           
  Branches                               ?        0           
==============================================================
  Hits                                   ?    33514           
  Misses                                 ?     2478           
  Partials                               ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[trivikr]

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

[metcoder95]

Agree with the optimization of #5085 but I also consider that this improvement is worth it, as keeping the raw headers on the controller can enable other use cases around