Attempt to kill workers once per request

[jclem]

This proposed change attempts to kill workers once per request. The reason for this is that the QUIT signal tells a Unicorn worker to shut down gracefully. Previously, WorkerKiller.kill_self would send the QUIT signal, then tell the worker to loop until it died. This loop prevented graceful shutdowns of the worker.

My first solution for this was to either fork the process (or use a Thread) and call Process.kill :QUIT, worker_pid from there, but this solution seems more elegant. In general Unicorn workers are good about gracefully shutting down when they receive a QUIT (is there significant evidence of them not?), so this solution attempts to send a QUIT first, and then escalates from there on each following request.

It's a somewhat significant change and a version bump, so I thought I'd open it for discussion briefly before merging and releasing.

[markpundsack]

Is it going to work with apps that are otherwise not threadsafe?