Less flaky multimode tests

[larskuhtz]

This PR fixes several issues with multi node tests:

• Multinode tests can overwhelm machines when the number of nodes is larger than the number of available nodes cores. This PR limits the number of nodes to the number of available cores. It also provides the option to set the number nodes via an environment variable, which is useful in containerized environments (where the number of available cores in the container may not match the number of physical cores) and in noise virtualized environments like GitHub runners.
• Fix the logging of the tests by routing logging output to the tasty progress loggers. This also fixes counting for the test statistics.
• Some adjustments to the node configuration.
• Some code cleanups.
• Reenable multinode tests in CI

At the moment replay tests are still somewhat flaky. We can either disable just those or try to further improve stability in the context of Github runners.

We should also do the following in CI:

1. Run multinode tests with larger numbers of nodes on more powerful hardware from time to time.
2. Run multinode tests on limited hardware with relaxed efficiency bounds while focusing on soundness.