Use `unsafeEncode` for parsing requests

I think we can make `RequestParser` much faster by not parsing `zhttp.http.Request` instead using `io.netty.handler.codec.http.HttpRequest`. You can access it via `unsafeEncode` method on `Request`.

This will dramatically improve performance, because -
1. For each real request parsing is done over `HttpRequest` which can reduce the overall number of allocations.
2. We can internally optimize the server to not create new `Request` at all when we are using this API :)

Start by locating RequestParser and the Request API, then read how unsafeEncode exposes the underlying HttpRequest. Trace the existing request-parsing path and its tests, if present; done means parsing uses the intended Netty request representation without changing request behavior and demonstrates the claimed allocation or performance improvement.

Written by the indexing model from the issue text.