Add background async health check for unhealthy pooled resources

[mdedetrich]

Resolves: #285

I did my best to take into account the concerns raised in r2dbc/r2dbc-pool#187 and I also ported a relevant test from HikariCP just to be doubly sure that everything works as expected.

The currently existing evictionPredicate wasn't suitable since it had a BiPredicate signature which is synchronous/non blocking so a new one had to be made (i.e. r2dbc-pool needs to do async requests against databases). The implementation works on the currently existing SimpleDequePool by adding health check ticks, making sure that while a resource is being checked it cannot be acquired (as well as other edge cases).

The expected configuration knobs have been added, I also added a TODO// to remind ourselves to remove unnecessary overrides when v2 gets released.

While working on this I have found a race condition which while it may not be problematic before, is now due to the newly added healthcheck i.e., previously there was

while (!e.isEmpty()) {
    QueuePooledRef<POOLABLE> ref = e.poll();
    if (ref.markDestroy()) {   // <- NPE here
        ...
    }
}

Individually e.isEmpty() and e.poll() are atomic (where e is a ConcurrentLinkedDeque) but combined in this way they are not, another thread could remove the last element and at the same time e.poll is accessed and you would get null (causing NPE) and this case happens with the PR's new healthCheckInBackground() (which runs in another thread) and it could have also happened with another future feature where another thread would be accessing e.