Optimize MTR resourcing to reduce failures of tests with replication setups

Feel free to add suggestions, comments, etc...

Some rpl (and replication-required binlog*) test failures on BB are due to not enough resources (e.g. MDEV-12906, MDEV-28986). This "noise" leads to unnecessary time spent looking at these fails.

As a replication setup requires another server instance to be started, this puts further strain on test runs that allocate a certain number of workers. This could be alleviated by altering the number of workers used when such a test is run, by doing the following:
If an rpl test is to be run, wait for TWO workers to be available (or schedule two workers only for one of them to sit idle) - presumably a change in the MTR script is required.

A future follow-up task could be:
Parse the preamble of a test to extract its rpl_topology and see how many MTR workers should be made occupied.