Revise integration tests to improve the total execution time

[dimakr]

Due to the growing number of integration tests we started hitting the situation when the tests are interrupted in CI due to timeout, e.g. https://jenkins.scylladb.com/job/sct-github-PRs-scan/job/scylla-cluster-tests/job/PR-11656/10/pipeline-overview/.

We need to revise and refactor the tests to improve their effectiveness (e.g. create a shared test setup for the whole test module, not for every test, where possible, etc.).
Maybe some of tests can be disabled as no longer relevant.

[fruch]

@pehala maybe you can do that while you do it on the unittest :)

[pehala]

Shared setup is not as big of an issue as I initially thought. Setting up scylla docker takes only a few seconds. With significant rewrite we could reduce the amount of clusters we spin up, but since we need to sometimes modify the cluster, it is not any easy fix and with the actual time it takes to spin it up, I do not think it is worth it.

My investigation has found that (so far) the biggest contributor to the actual time are the stress durations themselves. We use 1minute as default and it simply adds up due to the amount of tests we do.

I think the biggest gain could be had by either reducing the 1minute to 30sec or something similar, or by simply paralleling the tests (this will prove difficult as could be seen in #11726)