Support unlocking deadlocks with setnx/getset

[miyagawa]

This patch adds unlocking behavior when a worker somehow gets dead while obtaining a lock without cleaning it up and causes deadlocks. The code uses a design pattern used in http://redis.io/commands/setnx using SETNX and GETSET, to avoid the race condition.

To make it work i changed the value of lock to be an integer (UNIX timestamp) rather than a time string as it used to be.

This addresses #1.

[miyagawa]

The commits also bump up the resque dependency to work with the current release of resque. I've been running it on production for a couple of days now and see no problems with the current resque.

[miyagawa]

FWIW, 5 days is awfully long for a deadlock situation - I override that by defining our own def lock_timeout but that might not be obvious without looking at the source code. Might be worth adding a documentation?

[wallace]

Excellent, thank you!

[miyagawa]

There's some race condition reported in redis-mutex, which uses the same SETNX pattern: kenn/redis-mutex#5

I will need to see how they can get the issue fixed and if that can be applied to your gem too :)

[wallace]

Hm, I may have time to tackle this later this week.