Implements Sum,sum_checked,min,max,is Distict,inverse for REE.

[rich-t-kid-datadog]

Which issue does this PR close?

This PR works towards closing the larger REE Epic

Rationale for this change

Add operations onto the REE datatype such as

• Sum
• Sum_checked
• IS DISTINCT FROM
• Max/Min
• Distinct (Arrow already handles this apparently)

What changes are included in this PR?

Allows for REE columns to be used for the previously mentioned functions correctly and efficently.

Are these changes tested?

Yes, comprehensive tests have been added in arrow-ord/src/cmp.rs and arrow-arith/src/aggregate.rs:

• Basic functionality tests: Same values, different values, mixed scenarios
• Edge case tests: Empty arrays, single runs, all nulls, mixed nulls and values
• Data type tests: Float64 and Timestamp types
• Both operations: distinct and not_distinct operations
• The tests verify that REE distinct operations work correctly without array expansion and handle null values properly.

Are there any user-facing changes?

Performance improvement: REE distinct operations are now much faster for datasets with repeated values
No API changes: Existing distinct() and not_distinct() functions work the same way but are now more efficient for REE arrays
No breaking changes: All existing functionality is preserved

If there are user-facing changes then we may require documentation to be updated before approving the PR.

If there are any breaking changes to public APIs, please call them out.

[Jefffrey]

Just want to checkup on this PR, if it's still being worked on? Looking to see if we can get REE support pushed along in arrow-rs

[brunal]

According to #3520 (comment), the work was done while @rich-t-kid-datadog was an intern, and he's not available anymore. I can pick it up -- which should mostly be resolving merge conflicts.

[Jefffrey]

That sounds good, I'm willing to help review this to get it along

[brunal]

This isn't ready as-is:

• aggregate works, but is inefficient for min-max: it goes through all values (cost O(n*log(n)) for n values), when it could just look at the run-ends child array (cost O(# runs) < O(n))
• cmp only implements distinct, and is inefficient as well (binary searches for logical->physical index)

I'll send a first PR just for aggregate, and later a second PR for cmp.

[brunal]

Aggregate work done in #9409. In particular, it takes care of sliced Ree arrays.