fix(ext/node): throw ERR_INVALID_ARG_VALUE for odd-length headers array in http.ServerResp - #33820

LGTM. Tiny semantic fix — odd-length headers array passed to ServerResponse.writeHead() now throws ERR_INVALID_ARG_VALUE instead of ERR_INVALID_ARG_TYPE, matching Node's behavior.

What I verified

Error class change (ext/node/polyfills/_http_server.js:417): ERR_INVALID_ARG_TYPE("headers", "Array with even length", obj) → ERR_INVALID_ARG_VALUE("headers", obj). The semantic distinction is correct here — the argument IS an array (right type), but the value is malformed (odd length). Node's convention reserves ERR_INVALID_ARG_TYPE for cross-type mismatches and uses ERR_INVALID_ARG_VALUE for in-type malformations. ✓

Upstream test alignment (tests/node_compat/runner/suite/test/parallel/test-http-write-head.js): the upstream test asserts:

assert.throws(() => {
  res.writeHead(200, ['invalid', 'headers', 'args']);  // 3 elements, odd
}, { code: 'ERR_INVALID_ARG_VALUE' });

This PR's emission matches exactly. ✓

Test enablement (tests/node_compat/config.jsonc:1644): parallel/test-http-write-head.js added in alphabetical order between test-http-write-head-after-set-header.js and test-http-zero-length-write.js. ✓

CI

All test node_compat shards (debug + release × all 6 platforms × 3 shards = 36 jobs) pass. The newly-enabled test-http-write-head.js would land in one of the shards on every platform — all green confirms the change works end-to-end.

91 success, 8 in-flight (release linux-x86_64 integration + specs + unit + wpt + test specs debug windows-aarch64), 1 skipped, 0 fail. The remaining shards aren't in the _http_server.js blast radius.