Allow for reusing TCP connections

[jgaskins]

This eliminates TCP handshakes, TCP slow start, and potentially TLS negotiation for subsequent requests.

The max_idle_pool_size is just a starter value. I normally parse it from the URL the way the crystal-lang/crystal-db shard does, but I think it makes sense to discuss that as a feature first.

[stakach]

thanks for this @jgaskins
I do like the idea of this however I am a little concerned by the possibility of the clients in the pool becoming disconnected and the pool not recovering
crystal-lang/crystal-db#154 (comment)

i think we'd need to implement retries and make sure to catch errors and raise DB::ConnectionLost to ensure things are cleaned up properly

[jgaskins]

@stakach I thought HTTP::Client instances self-healing in those cases. Is that not true?

[stakach]

Looks like that was added recently, should drop in crystal 1.2 crystal-lang/crystal#11088

I've been having to deal with this manually by detecting closed sockets and creating new HTTP clients as required
https://github.com/PlaceOS/driver/blob/master/src/placeos-driver/transport/http_proxy.cr#L16

[jgaskins]

Oh interesting. I didn’t realize that hadn’t been implemented yet. I could’ve sworn I’d seen connections get closed and HTTP::Client would reconnect.

Ironically, I forgot to set Connection: keep-alive so it wouldn’t actually reuse the connection in its current state. 🙃 So it looks like there are a few things I need to update here.