Upon joining wsrep_order_and_check_continuity() can miss to commit some transactions

Discovered when testing the fix for MDEV-38147:

On the joiner, the wsrep XID continuity check (handler.cc:wsrep_order_and_check_continuity()) recovers a contiguous run of prepared wsrep XIDs starting at the SE checkpoint and stops at the first gap, potentially leaving some prepared transactions uncommitted.
The post-recovery dry-run ha_recover(0) (mysqld.cc:init_server_conponents()) then finds those leftover prepared transactions and aborts.

The non-contiguous prepared set in the mariabackup snapshot is almost certainly the the consequence of wsrep_slave_threads>0: parallel appliers preparing out of order.
and the per-transaction rotation at 4096 makes the snapshot land mid-gap. It only affects the joiner.