[REVIEW] Fast path when possible for non numeric aggregation

[VibhuJawa]

This PR adds fast path when possible for non numeric aggregation. The PR speeds up aggregations by avoiding the apply based path here.

Motivating Example

%%time
df =  pd.DataFrame({'grp_col': np.arange(0,100_000).repeat(3),
                    'p_date': [np.datetime64('2019-08', 'D'),np.datetime64('2019-07', 'D'),np.datetime64('2018-07', 'D')]*100_000,
                     })

df = dd.from_pandas(df, npartitions=4)
dc.create_table('table_1', df)

query = '''
        select 
            MIN(p_date) as f_date
        from 
            table_1
        group by
            grp_col
        '''
res = dc.sql(query)
res.compute()

Mainline

CPU times: user 39.5 s, sys: 451 ms, total: 39.9 s
Wall time: 37.1 s

With this PR (236):

CPU times: user 3.84 s, sys: 180 ms, total: 4.02 s
Wall time: 1.58 s

[VibhuJawa]

The PR should be ready for an inital review.

[ayushdg]

The changes look great and have a huge performance gains for these specific cases. Added a few comments inline.

It would be great to include some test cases to test_groupby as well that add code coverage for this case.

[VibhuJawa]

The changes look great and have a huge performance gains for these specific cases. Added a few comments inline.

Thanks , have addressed all of your reviews, I think this is ready for another review.

It would be great to include some test cases to test_groupby as well that add code coverage for this case.

I have included test cases for StringArray here, ea48c7a in test_compatibility.py .

Adding similar tests was not possible due to not being able to test equality with datetime columns due to a pandas issue with testing equality b/w string and datetime columns. (Raised the issue here: pd.testing pandas-dev/pandas#43707) .

Added Datetime tests with commit 6598259

For test cases corresponding to cudf/dask_cudf , i think we should first let https://github.com/dask-contrib/dask-sql/pull/229/files land to prevent any merge conflicts, Happy to figure out a workaround if we would really like dask-cudf specific test cases.

[charlesbluca]

rerun tests

[codecov-commenter]

Codecov Report

Merging #236 (3eb6f4b) into main (767a23c) will decrease coverage by 0.11%.
The diff coverage is 85.71%.

@@            Coverage Diff             @@
##             main     #236      +/-   ##
==========================================
- Coverage   99.61%   99.50%   -0.12%     
==========================================
  Files          64       64              
  Lines        2608     2622      +14     
  Branches      367      372       +5     
==========================================
+ Hits         2598     2609      +11     
- Misses          8        9       +1     
- Partials        2        4       +2     

Impacted Files	Coverage Δ
dask_sql/physical/rel/logical/aggregate.py	97.87% <85.71%> (-2.13%)	⬇️
dask_sql/physical/rex/core/call.py	100.00% <0.00%> (ø)

---

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 767a23c...3eb6f4b. Read the comment docs.

[galipremsagar]

rerun tests

[charlesbluca]

LGTM; thanks for the work here @VibhuJawa 😄