Move error handling to ensure loop continues after child process management

[macournoyer]

If a process dies right before we call kill! here:

async-container/lib/async/container/forked.rb

Line 249 in 2826b6e

self.kill!

, the resulting ::Process.kill(:KILL, @pid) will throw an error (Errno::ESRCH (No such process)), and the run loop will exit.

We do want to keep the loop running.

So this PR removes the re-raise, and move the rescue higher up so we count it as a failure.

[ioquatix]

In principle, I'm okay trying out this change.

However:

1. Killing the same process multiple times is okay. It's only when it's reaped it goes away (Process.wait).
2. The controller process should be the only one to reap children, otherwise it would be extremely problematic (think about if the pid was reused).
3. Since we are unsure of exactly what is failing, we might enter into a bad state. Even if we are limping along, it may ultimately be better to crash.

However, in the interest of robustness, and until we can establish the actual source of the issue, let's try this out.