chore: add better benchmarks

[BlankParticle]

Added a better benchmarks with extra cases inspired by srvx's nodejs compatibly suite, its also using oha instead of bombardier as srvx does. It uses more runs and measures in random order to account for random noise/activity, a full run takes about 10 minutes, --update automatically update the benchmark readme, there is a command provided to quickly smoke test the benchmark. We compare @hono/node-server (npm and local) and srvx against a baseline node:http version to measure overhead of the fetch servers in various scenarios.

---

Disclosure: These changes were created by GPT 5.6 Sol, by looking at hono node-server and srvx's repo under my supervision

[yusukebe]

Hey @BlankParticle

This is great! Can we also have a single ranked table like srvx has? Something like "one number per server, fastest first". Like the below:

[BlankParticle]

hello @yusukebe,
looks like srvx uses json round trip as the benchmark, i have added the section and it compares against node http as the baseline

[yusukebe]

LGTM!

[yusukebe]

looks like srvx uses json round trip as the benchmark, i have added the section and it compares against node http as the baseline

I see! Looks good. Thanks @BlankParticle