fix(node:http): emit headers passed via res.writeHead (#2132)

[proggeramlug]

Summary

Fixes a concrete byte-level diff from #2132: headers set via res.writeHead(...) were silently dropped from the http(s) server response.

writeHead's headers argument was typed NA_STR in the native dispatch table (native_table/http.rs), so when a user passed a headers object it was coerced to the literal string "[object Object]" — never valid JSON — and the runtime discarded it. The writeHead(status, statusMessage, headers) overload also couldn't be distinguished from writeHead(status, headers) because both arg slots were fixed strings.

Before / after (raw response bytes vs Node)

res.writeHead(200, { "Content-Type": "text/plain" }); res.end("Hello")

before: HTTP/1.1 200 OK\r\ncontent-length: 5\r\nconnection: close\r\ndate: ...\r\n\r\nHello   ← Content-Type dropped
after:  HTTP/1.1 200 OK\r\ncontent-length: 5\r\ncontent-type: text/plain\r\nconnection: close\r\ndate: ...\r\n\r\nHello

All overload forms now emit their headers: writeHead(status), writeHead(status, headersObj), and writeHead(status, statusMessage, headersObj).

How

• native_table/http.rs: writeHead args [NA_F64, NA_STR, NA_STR] → [NA_F64, NA_JSV, NA_JSV] so arg2/arg3 reach the runtime as raw NaN-boxed JSValues. The codegen extern decl was already [I64, DOUBLE, I64, I64], so this is ABI-compatible.
• response.rs::js_node_http_res_write_head: resolve the overloads in the runtime — is_string() (STRING_TAG) ⇒ statusMessage, is_pointer() (POINTER_TAG heap object) ⇒ bulk headers. Headers objects are serialized with js_json_stringify and merged via a new apply_headers_json helper (lowercase lookup key, original case retained for getHeaderNames()).

Scope

This fixes the dropped-header class of #2132's diffs. The remaining #2132 differences — lowercase header casing, header ordering, custom reason-phrase (202 Accepted!), and chunked-vs-content-length framing — are all serialized by hyper (which lowercases names, controls ordering, and derives canonical reason phrases) and are out of scope for this change; they'd need a separate response-writer change. The issue explicitly invites such per-cluster sub-fixes.

Test plan

• New unit tests in response.rs for apply_headers_json (case preservation, non-string value stringification, empty/sentinel handling).
• cargo test -p perry-ext-http-server — 17 passed.
• Verified end-to-end against Node v25 by capturing raw response bytes from a Perry http server through a raw TCP client for setHeader and all three writeHead overloads.