Support categorical splits in in TreeExplainer

[hcho3]

Stacked on top of #4447. Do not merge until #4447 is merged first.

If you'd like to review before #4447 is merged, look at the net diff at hcho3#2.

• Test XGBoost models with categorical splits
• Test LightGBM models with categorical splits

[RAMitchell]

Hypothesis is already in the rapids environment, although this is maybe not the same as us being allowed to use it :)

I wrote almost all my tests for #4492 in hypothesis.

[dantegd]

@hcho3 Rory's advice is correct, hypothesis already being used by cuDF, so it is already in our testing environments (and developer envs for that matter), so it should be fine to use in pytests

[hcho3]

@RAMitchell I agree that property-based testing is a good idea. Given that code freeze is in 3 days, let me add hypothesis based testing in a follow-up PR targeting 22.04.

[RAMitchell]

Good idea, it's not urgent.

[ajschmidt8]

Removing ops-codeowners from the required reviews since it doesn't seem there are any file changes that we're responsible for. Feel free to add us back if necessary.

[hcho3]

@RAMitchell @wphicks Can you take another quick look? I suggest that we merge this as long as there isn't any big blocker. We can address remaining issues in the following release, in which we'd move TreeExplainer out of experimental namespace.

[hcho3]

Rerun tests

[dantegd]

rerun tests

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (branch-22.02@3ccf77f). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-22.02    #4473   +/-   ##
===============================================
  Coverage                ?   85.69%           
===============================================
  Files                   ?      236           
  Lines                   ?    19338           
  Branches                ?        0           
===============================================
  Hits                    ?    16572           
  Misses                  ?     2766           
  Partials                ?        0           

Flag	Coverage Δ
dask	46.52% <0.00%> (?)
non-dask	78.59% <0.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3ccf77f...f97106a. Read the comment docs.

[dantegd]

@RAMitchell will go ahead and merge this PR so that it makes it to release 22.02, @hcho3 could you address any pending comments and further testing in a follow up PR? Thanks!

[dantegd]

@gpucibot merge