Write design for fault domain based replacements

[johscheuer]

What would you like to be added/changed?

The current automatic replacement implementation has the limitation that it only works on a per process group level. That means that the operator is not efficient at replacing multiple process groups when a host or a fault domain fails. Instead of concurrently replacing those process groups the operator will replace them one by one. We should extend the current implementation to focus on fault domains (or hosts) instead of single process groups.