[ci] Default to n=2 for test parallelism (#12376)

This decreases the test times for most of the tests except a few that did not run under pytest-xdist with 2 worker nodes. This also doesn't decrease overall runtime since CI is still bottlenecked on other jobs. However, this could lead to savings in compute which makes CI more sustainable so this is still worthwhile, though we should revert this if we start seeing "weird" errors like OOMs more often.

Author: driazati