Fix construction of nested structs with EMPTY child

[shwina]

This PR adds the ability to construct a structs columns with a parent null mask, but an EMPTY child. Prior to this PR, this would fail with a runtime error. This is because the parent null mask is superimposed on the EMPTY child, but EMPTY columns cannot contain null masks.

This PR also includes a small change in the Cython, where we map libcudf's EMPTY type to a int8 type on the Python side.

[codecov]

Codecov Report

Merging #10761 (c75dacf) into branch-22.06 (84f88ce) will decrease coverage by 0.08%.
The diff coverage is n/a.

@@               Coverage Diff                @@
##           branch-22.06   #10761      +/-   ##
================================================
- Coverage         86.40%   86.32%   -0.09%     
================================================
  Files               143      144       +1     
  Lines             22444    22656     +212     
================================================
+ Hits              19393    19558     +165     
- Misses             3051     3098      +47     

Impacted Files	Coverage Δ
python/cudf/cudf/comm/gpuarrow.py	79.76% <ø> (ø)
python/cudf/cudf/core/column/string.py	88.78% <ø> (-0.31%)	⬇️
python/cudf/cudf/core/frame.py	93.41% <ø> (ø)
python/cudf/cudf/core/series.py	95.17% <ø> (+<0.01%)	⬆️
...thon/dask_cudf/dask_cudf/tests/test_distributed.py	18.86% <0.00%> (-67.93%)	⬇️
python/cudf/cudf/io/parquet.py	90.83% <0.00%> (-1.87%)	⬇️
python/dask_cudf/dask_cudf/backends.py	85.51% <0.00%> (-0.94%)	⬇️
python/cudf/cudf/core/column/categorical.py	89.37% <0.00%> (-0.61%)	⬇️
python/cudf/cudf/core/column/decimal.py	90.60% <0.00%> (-0.50%)	⬇️
python/cudf/cudf/core/column/lists.py	91.70% <0.00%> (-0.38%)	⬇️
... and 28 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 3c208a6...c75dacf. Read the comment docs.

[vuule]

Huh, I always assumed that EMPTY just means an invalid type.

[davidwendt]

Is there a better way to test this? Perhaps the pytest is enough for this change.

[shwina]

Is there a better way to test this? Perhaps the pytest is enough for this change.

This is an issue that exists independent of Python -- simply constructing a structs column or reading an Arrow array with the specified structure would fail prior to this PR.

[shwina]

rerun tests

[shwina]

@gpucibot merge