Rework tests

[AlekseySh]

No description provided.

[AlekseySh]

Changelog:

• Starting now we will be running tests for all the supported pythons, instead of only for 3.8 to have better coverage. We will run short_tests for pull_request and long tests for push requests. Tests are running outside of docker, on the CI Ubuntu machine.

• To avoid excessive checkpoints downloading from Zoo for every Python version, this test is executed only when a special global env is provided: DOWNLOAD_ZOO_IN_TESTS=yes. So, we provide this env only for one of the Python versions from the whole list.

• We keep all_tests running in docker only for one Python version and only for push requests. It's done to avoid the need to maintain ~3-5 docker containers for every Python version and to save the time needed for every docker build.

• Marked a number of tests as long to make short_tests even shorter (not all of them are really "long", but it seems unlikely that we need to run all of them on every commit)

• Requirements needed for testing (pre-commit and pytest) are in one place and re-used in workflows & dockerfile

[DaloroAT]

What tests are not long? 😁