Improve pipeline speed and abort early

[ernestum]

Problem

In the last month we doubled our median pipeline runtime and we also spend a lot of CirlceCI credits on failed pipelines

Solution

1. Make the pipeline faster
   • use ruff instead of flake8
   • stop using custom docker images (should load faster since default images are still in the cache of the runners)
   • weed out really slow tests (the new SQIL tests are slooow)
   • Use the parallel test runs feature of CircleCI
2. Make the pipeline fail earlier so we spend less credits on failed pipelines
3. Finally remove the windows pipeline (I assume it has the least amount of users and causes the highest amount of pipeline errors). I think windows users reporting issues when they come up for them might be enough.
4. Only run tests that cover changed sections of the code
   • Try pytest-incremental
   • Try pytest-testmon
   • Codecov is offering something along the lines
   • Split up the repo (e.g. imitation-core with shared stuff, imitation-<algo> one for each algo and imitation which is a meta-package and CLI to all the algos). Then each push should only run the tests for the algo/part that actually changed

Possible alternative solutions

• Find ways to change perception of time. Then waiting for the pipeline might become less annoying.
• Stop caring for money spent on compute and waiting humans.