Automatic performance benchmarking

[juhoinkinen]

We could utilize pytest-benchmark, a plugin for pytest, to monitor the performance of some operations.

Performance regressions affecting Annif can occur not only by changes in Annif code, but also due to changes in dependencies, and this is not always easy to notice. In NLTK there have been two regressions affecting Annif: nltk/nltk#3013, nltk/nltk#3299.

Currently there is a check for the CLI startup time (fast CLI startup is important for tab completions to be usable), which is ran in GH Actions.

There could be unit tests for the performance of the train and suggest operations of backends (or at least some important backends) that would be run in the CICD pipeline with the GitHub Action "Continuous Benchmark".