[#21878] extensive test for multi-dictionary column group bys

[Rich-T-kid]

Which issue does this PR close?

works towards closing #21878. But also adds test coverage for existing code that wasnt there before. There were no test for dictionaries specifically for groupValueRows

Rationale for this change

There was a lack of test that focused on dictionary arrays in GroupValueRows. I also would like to keep the test separate from the implementation (for #21878) so that its easier to review.

What changes are included in this PR?

• Adds a dedicated test module (multi_group_by/dictionary.rs) covering correctness of multi-column dictionary GROUP BY using the GroupValuesRows path
• Tests are organized into three modules: general_test (uniqueness, null semantics, multi-type composition), dictionary_encoding (key-index shuffling, bloated/non-normalized dictionaries, mixed key types), and streaming (EmitTo::First(n) renumbering and EmitTo::All reingest)
• Adds two SQL logic tests to dictionary.slt covering multi-column dictionary GROUP BY with mixed key/value types (Dict<Int8,Utf8>, Dict<Int32,LargeUtf8>, Dict<Int8,Utf8View>). these are the first SLT tests where the grouping keys themselves are dictionary columns

Are these changes tested?

The changes are test 😆

Are there any user-facing changes?

No

[Rich-T-kid]

I tried to make each test function as declarative as possible.

[jayshrivastava]

Looks great!

[Rich-T-kid]

@2010YOUY01 I know you have a lot of experience with aggregation operator in datafusion . Do you mind taking a look at this smaller PR, when you get a chance?

[2010YOUY01]

I don't have enough context on the dictionary type support right now, but this is pretty cool. I hope to find some time to look into it later.

Here is a bit of context I can share: I assume we would need to specialize a GroupValues implementation for dictionary types? Right now, we are going through a larger refactor to change the GroupValues contract for better memory efficiency, so I plan to look into dictionary support after the initial blocked-memory patch lands, to save some efforts. I'll @ you when there is progress.

TL;DR: today, GroupValues allocates memory as a single contiguous vec for all groups. We want to change that to fixed-size blocks. The issue below has more details:

• Improve aggregate performance with adaptive sizing in accumulators / avoiding reallocations in accumulators #7065

[Rich-T-kid]

Thx, I'll defiantly take a look at the issue when I get a chance!

[Rich-T-kid]

I do think these test still provide value, @kumarUjjawal what are your thoughts? Should I close this PR as well?

[kumarUjjawal]

I do think these test still provide value, @kumarUjjawal what are your thoughts? Should I close this PR as well?

In my opinion since you are going to be working on more centralised dictionaries implementation, these tests should wait till we have the structure or you can port these to the new prs.