improve unrecoverable resource registration retrial and recoverer implementation

Unrecoverable resource registration retrial implementation currently is very poor as it destroys & unregisters the pool before re-creating and re-registering a new one when a recovery error happens. This can be disastrous if the user manually created the pool and kept a reference to it.

The background recoverer suffers from potential race conditions and a poor default behavior (it's disabled by default for no good reason for instance).

This task is about merging the unrecoverable resource registration retrial implementation in the recoverer while at the same time re-implementing the recoverer to make it race conditions proof and carefully choosing new default values. Both can be done with very little code change.