AWS Aurora's replication is more akin to Always On Availability Group. The primary pushes storage changes to other replicas. You don't get to make schema changes on the subscriber.
More details are available in several re:Invent deep dive sessions.
Those are probably bogus.
If you have more than a few dozen connections, they will be stumbling over each other and everyone will progress very slowly. Keep the number of threads to a realistic number.
autocommit=0 should be outlawed. It begs the programmer to forget to issue COMMIT. Without COMMIT, a disconnect (or certain other things) will lead to a ...