This better models the way MRI uses the Mutex. Rather than
directly doing the unlock, sleep, lock steps, this version just
adds the thread to a listeners queue and asks the mutex to sleep.
This allows us to reuse the sleep logic already in Mutex, which
now properly defers to RubyThread methods to do its own unlock,
sleep, lock cycle. This centralizes all of the key Mutex behaviors
in RubyThread so they can be properly boxed with thread status
changes like "sleep".

This also fixes jruby#5476 by restoring full interruptibility to Mutex
contended lock calls. Both interrupts from Java and from Ruby will
wake the thread. This returns us to matching MRI's Mutex lock
interrupt behavior.

* Arity split #wait all the way through to avoid nil coercion
  error in arity-split Mutex#sleep.
* Remove waiters while interrupting in #broadcast.

* Properly unify Mutex#sleep arities to match CRuby behavior.
* Unify ConditionVariable#wait_ruby logic again.