AMQ-9693: avoid synchronized blocks

[jeanouii]

This PR fixes a couple of issues in the RA.

• move the running flag to a non-blocking approach (remove synchronized on a mutex and use AtomicBoolean)
• update the running to false if we can shedule a task in the worker
• move the listener initialization to the #start which is functionally better and avoid stacking messages in the session while it's not started
• avoid pulling a stale (or not running) session from the pool if we can find otherwise

[jbonofre]

I guess the problem is here as the runningFlag is not thread safe (and we can have several session), right ?

[jeanouii]

This is a bit of a trick, because this avoids the lock but does not fix yet the initial issue in my opinion. There is an underlying bug revealed by this deadlock allowing another thread to be using this session to dispatch unconsumed messages from a stale session (removeFromPool).

Trying to properly understand it, so I can push a more complete fix

[jeanouii]

So the trick can happen when there is no more session in the pool, then ActiveMQ pulls an active session leading to a deadlock as I have observed. With the AtomicInterger, the integrity is preserved without any locks.