pytest collecting phase is slow

[albertz]

Running some simple selected test takes a long time. Most of that time is spent in the collecting phase.

That's very annoying for debugging. Also, I don't quite understand why it is so slow. But you find quite a number of related reports, e.g. this, this, this. They mostly suggest to restrict the paths which are searched for the tests. We already do that, and that doesn't really seem to have much impact. Although, when running a single specific test, I also don't really understand why it needs to search through all the other tests.

E.g. when running python -mpytest test_rf_array.py::test_gather_3d_embed with strace -e trace=file:

11:35:06 execve(...)
...
11:36:08
collected 1 item
...
11:36:09 +++ exited with 0 +++

Of course, this imports torch, tensorflow, numpy, and some other libs, and this already takes some time, but this should be only 10 secs or so.

We should better understand what takes so long, e.g. by profiling/tracing this. And then maybe we can fix this.

(cc @NeoLegends)