Solution: remove and add where needed

Solution: use unity assertions instead

Solution: do it

Solution: allow for a range between the minimum and maximum expected.
With 2 queues, that's between 20 and 40 messages depending on the
scheduling of the 3 threads involved.

Solution: use atomics to avoid issues on non-x86 architectures

Improve test reliability by: a) using XPUB in place of PUB to ensure we start publishing only after 1st subscriber has joined; b) accept both 2*HWM, 3*HWM and 4*HWM as TX/RX count of messages

Add new HWM tests and more detailed documentation