perf(aot): cold-start, RSS and binary-size benchmark for compiled servers

Problem

Nothing measures the server claims. bench/ has lexer/parser benches and the TS checker comparison; bench/vs_zig.sh is stale (still uses ION_ variables). The roadmap's cold-start and memory goals are unchecked. (New.)

Evidence (origin/main @ d4b219507)

• bench/vs_zig.sh:194
• docs/docs/ROADMAP-WEB-COMPETITIVE.md:750-752

Scope

• A benchmark that builds the same minimal HTTP server with: home build --compile (with and without --bytecode), a natively compiled Home server, bun build --compile, Node SEA, deno compile, and a Rust hyper/axum baseline.
• Measure time-to-first-response, peak RSS under a fixed load, requests/s and binary size.
• Fix or delete bench/vs_zig.sh.

Acceptance criteria

• Script checked into bench/ with a README.
• A CI job (or scheduled job) publishes the numbers.

Links

Depends on: #791, #793 Part of #814.