Improving Python DataType support for Struct and repr

[cjermain]

While working on #3413, I noticed that the Python representation of DataTypes has some issues. This PR fixes a few of them:

• Struct dtype could not be converted back to DataType::Struct, since previously it had an incomplete representation
• dtype_string_repr was not implemented yet, so it always failed to import

The __repr__ behavior is a bit odd, but much closer in this PR to the Rust API. This is because the class is being used as a singleton without being constructed -- type(dtype) == type. So for native types repr(dtype) is actually type(dtype).__repr__(dtype) which gives strings like "<class 'polars.datatypes.Utf8'>". I left the DataType.string_repr method, but made it more clear that as a classmethod this uses cls not self. I think this could be improved by having an instantiated singleton that the class provides on the new constructor. This can be a follow up.

[cjermain]

I'm not seeing the tests fail locally. I'm wondering if the test_arr_unique test is intermittent:
https://github.com/pola-rs/polars/runs/6543827625?check_suite_focus=true#step:9:514

[codecov-commenter]

Codecov Report

Merging #3471 (d45d6d6) into master (f14570a) will decrease coverage by 0.00%.
The diff coverage is 62.96%.

@@            Coverage Diff             @@
##           master    #3471      +/-   ##
==========================================
- Coverage   77.61%   77.61%   -0.01%     
==========================================
  Files         408      408              
  Lines       69847    69876      +29     
==========================================
+ Hits        54214    54234      +20     
- Misses      15633    15642       +9     

Impacted Files	Coverage Δ
py-polars/polars/__init__.py	100.00% <ø> (ø)
py-polars/src/lib.rs	93.51% <16.66%> (-1.27%)	⬇️
py-polars/src/conversion.rs	78.27% <64.28%> (-2.72%)	⬇️
py-polars/polars/datatypes.py	92.00% <72.22%> (-2.65%)	⬇️
polars/polars-core/src/datatypes.rs	69.57% <100.00%> (-0.20%)	⬇️
...olars/polars-core/src/frame/groupby/into_groups.rs	54.21% <0.00%> (-0.37%)	⬇️
...ars/polars-core/src/chunked_array/ops/aggregate.rs	94.30% <0.00%> (+0.09%)	⬆️
polars/polars-core/src/series/mod.rs	57.85% <0.00%> (+0.13%)	⬆️
... and 9 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 f14570a...d45d6d6. Read the comment docs.

[ritchie46]

Could you do a rebase on master? That should fix it.

[ritchie46]

Thanks for this. I agree it's something we can improve because of the nested dtypes.