Unfortunately, elastic#42791 destabilized SLM tests because those tests use
rate limiting the snapshot write rate to a very low value globally.
Now that the various files in a snapshot get uploaded in parallel
this can lead to a few threads in parallel way overshooting the low
value throughput value used by the rate limiter and then making it
wait for minutes which times out the tests that then try to abort
the snapshot (see elastic#21759 for details, aborting a snapshot only
happens when writing bytes to the repository).

For now the old behavior of the test from before my changes can
be restored by moving to a single threaded snapshot pool but
we should find a better way of testing the SLM behaviour here in
a follow-up.