Connections leaks in server about httpaf

[paurkedal]

I investigated this a bit further, and the remaining connections handlers after wrk exists seems to be stuck in read. This is likely the issue described e.g. in Detection of Half-Open (Dropped) Connections, which explicitly mentions timeout as the solution for HTTP servers which support persistent connections. So, it's not a bug.

For a production-ready server we would need a timeout. I don't know Async well, but Lwt this can be easily done by joining the handle thread with an Lwt_unix.timeout, but this is not fully correct, since a big transfer could be cancelled even though it's still streaming. So, maybe the handler constructors should take a ~idle_timout or ~read_idle_timeout parameter? (Same semantics, different name suggestions.)

[seliopou]

Thanks for the report, Closing in favor of #86.