Implement DataFrame.eval using libcudf ASTs

[vyasr]

This PR exposes libcudf's expression parsing functionality in cudf and uses it to implement DataFrame.eval. The implementation is mostly feature-complete, but there are a few limitations relative to the pandas API and a couple of gotchas around type casting. The implementation is reasonably performant, improving upon an equivalent df.apply even accounting for JIT-compilation overhead. This implementation provides a stepping stone to leveraging libcudf's AST implementation for more complex tasks in cudf such as conditional joins.

The most significant issue with the current implementation is the lack of casting between integral types, meaning that operations can only be performed between columns of the exact same dtype. For example, operations between int8 and int16 would fail. This becomes particularly problematic for constants e.g. df.eval('x+1'). The best paths to improve this are at the C++ level of the expression evaluation, so I think we'll have to live with this limitation for now if we want to move forward.

Resolves #9112

[codecov]

Codecov Report

Merging #8022 (11d81d6) into branch-22.06 (01d08af) will increase coverage by 0.07%.
The diff coverage is 93.77%.

❗ Current head 11d81d6 differs from pull request most recent head f009e5b. Consider uploading reports for the commit f009e5b to get more accurate results

@@               Coverage Diff                @@
##           branch-22.06    #8022      +/-   ##
================================================
+ Coverage         86.35%   86.42%   +0.07%     
================================================
  Files               142      143       +1     
  Lines             22335    22438     +103     
================================================
+ Hits              19287    19393     +106     
+ Misses             3048     3045       -3     

Impacted Files	Coverage Δ
python/cudf/cudf/_fuzz_testing/fuzzer.py	0.00% <0.00%> (ø)
python/cudf/cudf/_lib/__init__.py	100.00% <ø> (ø)
python/cudf/cudf/core/indexed_frame.py	91.70% <ø> (ø)
python/cudf/cudf/testing/dataset_generator.py	73.25% <ø> (ø)
python/cudf/cudf/testing/testing.py	81.69% <ø> (ø)
python/cudf/cudf/utils/utils.py	90.35% <ø> (+0.06%)	⬆️
python/dask_cudf/dask_cudf/io/orc.py	91.04% <ø> (ø)
python/cudf/cudf/core/column/numerical.py	96.17% <50.00%> (+0.29%)	⬆️
python/dask_cudf/dask_cudf/io/parquet.py	92.39% <81.08%> (-1.40%)	⬇️
python/cudf/cudf/core/_internals/expressions.py	92.85% <92.85%> (ø)
... and 24 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 8a4d1b2...f009e5b. Read the comment docs.

[harrism]

Moving to 21.10

[brandon-b-miller]

I have some thoughts about the mixed dtype issue, but they're probably a little out of scope of this PR. Let's follow up when you get a chance.

[bdice]

I have several comments that I would like to see addressed. For now, I am requesting changes. If you would like to punt all these comments/fixes to a future PR, I will approve and let you do that separately so that this large chunk of changes doesn't get stale.

[bdice]

A few minor comments but this looks great!

[brandon-b-miller]

I think all my comments were addressed here.

[vyasr]

@gpucibot merge