Implement keep_alive_timeout option

[robmueller]

There's a well known race condition with http keep alive connections. If the server closes the connection after the client has checked the socket is still connected, but before it has prepared the request to send it, then the client will receive an EPIPE error and fail the entire request when it tries to write the request to the socket.

This means clients always have to check for this case and decide what to do, which could be resend the request depending on when exactly the EPIPE occured and whether the request was idempotent or not.

Another work around for this is to make sure the client has a keepalive reuse timeout that's less than the server keepalive time. So for instance if the client has a 5 second timeout and the server has a 10 second timeout, then if it's been > 5 seconds since the connection was last re-used, the client will close it and open a new one, avoiding the case where the server would close the connection on it.

This setup can be very useful when using HTTP::Tiny to send requests to internal service endpoints and wanting as much as possible to avoid this edge case failure more.

[haarg]

This overall seems good to me.

One concern I have is that it only supports whole seconds. The existing timeout method will under the hood be implemented using select, so it will support finer granularity. Possibly this should be using Time::HiRes::time.

[robmueller]

I've updated the PR to use Time::HiRes iff keep_alive_timeout > 0, minimise classes loaded unless needed.