http: reduce per-request allocations in the server

[mcollina]

Removes allocations and repeated work the HTTP server performs for every request:

• Per-request bound functions: updateOutgoingData.bind() becomes one closure per connection reused by every response, and the per-request resOnFinish.bind() becomes a shared 'finish' listener that reaches everything through the connection state stored on the response.
• req.headers materialization: the requireHostHeader check (req.headers.host), the Expect check, and optimizeEmptyRequests' body-header check now scan rawHeaders (over the same range the headers getter exposes) instead of building the whole headers object — with a toLowerCase() per header name — for every request even when the application never reads req.headers.
• ServerResponse options object: cached per server; custom response classes keep receiving a fresh object since they may retain or mutate it.
• Status line: cached per status code when the reason phrase is the default (also skipping its character validation, which cannot fail for the built-in phrases).
• Date: header line: the complete line is cached alongside the existing one-second utcDate() cache.
• Keep-alive header pair: Connection: keep-alive\r\nKeep-Alive: timeout=N is rendered once per server settings instead of per response.
• Lenient-validation lookup: the 4-level option chain is computed once per message.

Responses are byte-identical to main (verified by diffing raw responses).

Benchmarks (i7-7700, server pinned to one core, wrk -t2 -c50 on separate physical cores, interleaved rounds, CPU from /proc/<pid>/stat):

workload	main	this PR	Δ
hello world, 1 request header	26.0 µs/req	24.2 µs/req	−6.9% CPU, +7.5% req/s (8/8 rounds)
hello world, 8 request headers	28.7 µs/req	25.3 µs/req	−11.8% CPU, +13% req/s (5/5 rounds)

Sampled heap allocation drops from 5206 to 4531 B/request (−13%); the req.headers build (~200 B + per-name toLowerCase per request) disappears from the profile entirely.

test/parallel/test-http-*, test-http2-*, test-stream-*, test-net-*, test-https-* and the http diagnostics-channel suites all pass (1171 tests).

Follow-up to #65732 and #65749.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TV8WVjiHJPR2hW4D6h1Xpr

[nodejs-github-bot]

Review requested:

• @nodejs/http
• @nodejs/net

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/76996/

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.15%. Comparing base (452c8c5) to head (ed051a3).
⚠️ Report is 133 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #65802      +/-   ##
==========================================
+ Coverage   90.05%   90.15%   +0.09%     
==========================================
  Files         754      769      +15     
  Lines      255727   261721    +5994     
  Branches    48312    49688    +1376     
==========================================
+ Hits       230286   235944    +5658     
- Misses      16564    16795     +231     
- Partials     8877     8982     +105     

Files with missing lines	Coverage Δ
lib/_http_incoming.js	97.99% <100.00%> (+0.05%)	⬆️
lib/_http_outgoing.js	97.93% <100.00%> (+0.02%)	⬆️
lib/_http_server.js	96.82% <100.00%> (+0.20%)	⬆️
lib/internal/http.js	92.59% <100.00%> (+0.25%)	⬆️

... and 158 files with indirect coverage changes

🚀 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.