Improve performance of `corr` function

[alamb]

Is your feature request related to a problem or challenge?

The corr function is used in the h20 benchmark and is quite slow.

https://github.com/apache/datafusion/blob/main/datafusion/functions-aggregate/src/correlation.rs

It would be great to make it faster so our performance in the H2o benchmark improves

Describe the solution you'd like

See details on #13548

1. Add a benchmark for corr function
2. Improve performance of corr function (likely by implementing GroupsAccumulator)

Describe alternatives you've considered

No response

Additional context

Used here in @MrPower's benchmark:

https://github.com/MrPowers/mrpowers-benchmarks/blob/0b586a0657d7f6cfd55d89508e15b95e79bd4010/benchmarks/datafusion_h2o_groupby_queries.py#L37

[2010YOUY01]

take