[unittest] supported condititonal testing based on env var

[FrankLeeeee]

What is the problem?

During development, we often need to temporarily skip some tests, either because they take too long or they are currently not working well. However, we can easily forget that these tests are in fact required to verify the correctness of each commit. To solve this problem, skipping a test should be controlled by a condition. Before we put up a pull request to the GitHub, we can set the condition flag locally and run a full range test without manually removing pytest.mark.skip in the code.

What does this PR do?

This PR added a util function run_on_environment_flag to colossalai.testing. This function receives a environment variable name and check if it is set to '1'. If so, this test is executed, otherwise skipped.

This PR also replaced pytest.mark.skip with run_on_environment_flag for the auto_parallel module testing.

Demo

• Run full-range testing without skipping.

AUTO_PARALLEL=1 pytest tests/test_auto_parallel/

• Run testing with some tests being skipped.

# if you do not set AUTO_PARALLEL, it is the same
AUTO_PARALLEL=0 pytest tests/test_auto_parallel/