Implement chunked TE

This can be tricky, because:

- If a response is set `Content-Length` header - we should not encode the body.
- If there neither `Content-Length` header nor `Transfer-Encoding: chunked` - we should encode the body. Hyper can do it easily.
- If there is `Transfer-Encoding: chunked` - we should firstly decode the body and to encode it again to avoid double-encoding.

Another issue is - there is some kind of accept logging. In the case of chunked TE it will be broken, as Hyper provides no way to notify when the last body chunk was written.

Start by tracing the HTTP response handling and Hyper integration, then inspect the accept logging path mentioned in the issue. Verify behavior for responses with Content-Length, without either header, and with Transfer-Encoding: chunked; done means bodies are encoded once and accept logging completes after the final chunk.

Written by the indexing model from the issue text.