Fix mindeps-distributed CI build to handle numpy/pandas not being installed

[jrbourbeau]

No description provided.

[jrbourbeau]

Okay, so this is an interesting one. On main we've started getting

FAILED dask/tests/test_layers.py::test_dataframe_cull_key_dependencies[_shuffle_op] - ImportError: Dask dataframe requirements are not installed.

Please either conda or pip install as follows:

  conda install dask                     # either conda install
  python -m pip install "dask[dataframe]" --upgrade  # or python -m pip install

in our mindep-distributed build due to numpy/pandas not being installed. Looking closer at test_dataframe_cull_key_dependencies, we are in fact missing a pytest.importorskip("dask.dataframe") (which has been added in this PR). Though I was confused by we weren't seeing this before. It turns out that previously we were skipping this test (and all the test in test_layers.py) due to the following issue when attempting to import distributed:

dask/tests/test_layers.py:5: could not import 'distributed': tlz.peekn

Which was recently fixed in dask/distributed#7309. So we had been skipping this test, but for the wrong reasons -- this PR makes it so we skip/run tests appropriately

cc @charlesbluca @ncclementi in case either of you have a moment to look at this

[jrbourbeau]

Will plan to merge this after CI finishes in order to get CI passing ASAP. Happy to handle any related changes in a follow-up PR

[rjzamora]

Thanks for looking into this @jrbourbeau ! CI failure seems like a known flake.

EDIT: Looks like my approval was a few seconds too late :)