Fix memory leak caused by async tracking busy threads

[mariokostelac]

Some printf debugging foo and reading the code helped to figure out that we were sporadically leaking Thread objects when processors died with an error.
https://github.com/phstc/shoryuken/blob/e83a68bf9d460fdd368a38208032074cde085ca2/lib/shoryuken/processor.rb#L15 calls real_thread (which pairs Thread and Processor in manager's namespace) asynchronously and it sometimes executes before processor_died is executed.

I've set up logs for every addition/removal of Manager.threads elements and state after the action. Worker I've used failed randomly, approx 33% of times.

Logs:

A: ADD 70141427912260 <-- worker starts
A: KEYS [70141427912260]
A: DONE 70141427912260 <-- worker done
A: DELETE 70141427912260 <-- thread deletion command
A: KEYS [] <-- thread deleted
A: ADD 70141427912260
A: KEYS [70141427912260]
A: DONE 70141427912260
A: DELETE 70141427912260
A: KEYS [] <-- threads empty
A: DIED 70141427912260 <-- worker died
A: DELETE 70141427912260 <-- delete Thread! but hey, it's not even added
A: DELETE 70141427912260 FAIL!
A: KEYS []
A: ADD 70141427912260 <-- oh ok, we're adding it here?! it will stay in forever
A: KEYS [70141427912260]
A: ADD 70141422012480
A: KEYS [70141427912260, 70141422012480]

Fixing #88.

One +1 for simplifying the core and getting rid of asynchronous stuff where it's not needed. We should be more deliberate in defining thing as actors and establishing async communication between them.