Fix TIMESTAMP / DATE scalars, add support for DATE column casting

[charlesbluca]

This is a continuation of #298 that also aims to implement CAST(... AS DATE) properly, as up until now that would work identically to CAST(... AS TIMESTAMP):

• Casts DATE / TIMESTAMP scalars to np.datetime64 instead of datetime.datetime for compatibility with Pandas; for whatever reason datetime.datetimes work in cuDF, which should probably be documented somewhere
• Casts DATE scalars to np.dtype("datetime64[D]") and TIMESTAMP scalars to np.dtype("datetime64[ns]"), so that DATE filters work as expected
• Adds special support for CAST (... AS DATE) using the datetime column's dt.date attribute; this currently doesn't work in dask-cuDF as it is dependent on the datetime accessor of cuDF, which doesn't have a date attribute
  • In general, this is a hacky workaround for truncating timestamps that is required because both Pandas and cuDF don't allow truncation through astype("datetime64[D]"); in Pandas this only applies for timezoned datetimes, and in cuDF this applies for all datetimes

Closes #296
Closes #298

[beckernick]

@charlesbluca is rapidsai/cudf#7880 in the critical path for this PR, or are there relatively straightforward workarounds?

[charlesbluca]

Need to look through the code again to make sure this is accurate, but IIRC another path forward here would be to add support for casting pandas timezoned datetime / cudf datetime columns to datetime64[D] - if that were the case, we would be able to truncate the timestamps to dates with a single astype call and avoid the use of the DatetimeAccessor.

[codecov-commenter]

Codecov Report

Merging #343 (e617a5c) into main (34fee74) will increase coverage by 0.16%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #343      +/-   ##
==========================================
+ Coverage   88.92%   89.09%   +0.16%     
==========================================
  Files          68       68              
  Lines        3333     3337       +4     
  Branches      651      653       +2     
==========================================
+ Hits         2964     2973       +9     
+ Misses        298      289       -9     
- Partials       71       75       +4     

Impacted Files	Coverage Δ
dask_sql/mappings.py	100.00% <100.00%> (ø)
dask_sql/physical/rex/core/call.py	99.12% <100.00%> (+<0.01%)	⬆️
dask_sql/_version.py	34.00% <0.00%> (+1.44%)	⬆️

---

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 34fee74...e617a5c. Read the comment docs.

[charlesbluca]

@ChrisJar I've checked and this should resolve #296 - if you get a chance, could you verify?

[ChrisJar]

@ChrisJar I've checked and this should resolve #296 - if you get a chance, could you verify?

I can confirm that this fixes all of the issues I was seeing. Thanks for working on this!