1. The issue is only reproduced in primary backup topology, There is no problem in the case of 0 backups. - Right, I could not reproduce the issue with 1 node per DC topology.
2. The issue started after upgrading from 10.2 to12.3 and blocking you from a release that was scheduled to the end of this month. Yes, correct.
3. Bootstrap is done on maintenance windows when system has issues, and should not take long. Correct, bootstrap is performed every time they to restart the system for any reason. It was not taking long on previous releases.
4. The problem occurs when bootstrap failed to replicate one of the entries, In that case, replication from the target back to the source is not active, till you are restarting the source and doing the bootstrap process again. Correct, sometimes they need to repeat the procedure several times until everything is in correct state.