http: improve performance by removing async_hooks

[JonasBa]

Reopening this from #57746.

[nodejs-github-bot]

Review requested:

• @nodejs/http
• @nodejs/net
• @nodejs/performance

[Flarna]

I marked this as semver major.
Removing async hooks for HTTP and as a result likely breaking AsyncLocalStore for HTTP requests will have significant impact to at least OpenTelemetry and other APM tools.

While I agree that HTTP parser itself is not async HTTP requests are. As a result I'm fine to the change of http parser itself but the async hooks functionality itself should preserved on HTTP request level otherwise then.

[Qard]

@Flarna have you verified that it's still relevant with the AsyncContextFrame-backed AsyncLocalStorage? I have a feeling it might not matter anymore, but not 100% certain about that. 🤔

[Flarna]

No, I haven't verified this. Basically this reverts fixes done a while ago (#27477 and #25094) to prepare HTTP for a better AsyncLocalStorage implementation.

Could be that AsyncContextFrame-backed AsyncLocalStorage solves this meanwhile, but it could be also that the move to AsyncContextFrame broke it already.
To my understanding AsyncContextFrame solution either follows promises/microtasks automatically inside v8, or it's told from outside. This happens for example in AsyncResource class on JS side or AsyncWrap class in C++. And exactly this AsyncWrap is removed here from HTTP parser.

In the end we don't know as there are no test (ref: #55712 - I thought I find some more time to add tests...).

[Flarna]

Created #58017 to add at least one more test for ALS + HTTP.

At the moment this branch doesn't build so couldn't do any local testing. As it is in draft state this is fine and there is no need to hurry anyway.