Register the TSTP signal handler before invoking the first job

[mdkent]

Fixes an issue where if the first job is a long running one, and you issue a SIGQUIT to the resque-pool-manager for a gracefully shutdown, you'll end up with your worker in a stopped state (T) as the trap() hasn't been set. eg:

root       941 85.5  0.7 392348 211224 ?       Sl+  18:34   0:08  \_ resque-pool-master[7bba15]: managing [949]
root       949  3.0  0.7 459940 200488 ?       Sl   18:34   0:00      \_ resque-1.26.0: Forked 957 at 1527705279
root       957  0.0  0.7 459940 199704 ?       Sl   18:34   0:00          \_ resque-1.26.0: Processing demo since 1527705279 [LongJob]

$ kill -QUIT 941

root       941 47.5  0.7 392348 211224 ?       Sl+  18:34   0:08  \_ resque-pool-master[7bba15]: managing [949]
root       949  0.3  0.7 459940 200488 ?       Sl   18:34   0:00      \_ resque-1.26.0: Forked 957 at 1527705279
root       957  0.0  0.7 459940 199704 ?       Tl   18:34   0:00          \_ resque-1.26.0: Processing demo since 1527705279 [LongJob]

Oops! It stays wedged like this indefinitely.

[nevans]

I had some issues (e.g. after_fork hooks not getting called when they should) when I tried to simply move hijack_fork above perform. I fixed the tests in #11, and fixed the TSTP bug (plus tests) in #13, by simply moving the trap (and nothing else) before the first perform.

[stulentsev]

@nevans Thanks for taking the time to contribute. Lemme have a coffee and review your PRs.

[stulentsev]

Fixed in #13