Implement meaningful Benchmarks that can be used as performance regression tests

We technically have benchmarks right now but those are not meaningful.
We should write benchmarks that reflect common use cases to be able to use them as regression tests before merging PRs.
It can still be decided if this should be integrated into CI.

Start by locating and reviewing the existing benchmarks in the repository, then identify which common Nemo use cases they should represent. Define how benchmark results will serve as regression tests and whether they belong in CI; the work is done when meaningful benchmark coverage and an agreed execution approach are established.

Written by the indexing model from the issue text.