colexec: implement vectorized table statistics collection

[yuzefovich]

Statistics from the tables are currently collected using a combination of row-execution sampler and sampleAggregator processors which have been introduced before we began implementing the vectorized engine. I believe that stats collection will benefit noticeably from the vectorized approach (similar benefits as in the other use cases - faster execution, better memory management). In particular, I'm hoping that it will alleviate issues like #54670.

[asubiotto]

@RaduBerinde I think the optimizer team needs to prioritize this for 21.1 given that #54670 is an easily reproducible OOM and we've seen a lot of OOMs related to wide rows lately. We're happy to offer guidance here.

[RaduBerinde]

@rytaft I thought we had hit OOMs with large rows before and had a workaround in place, do you remember the details? I think it was around dynamically adjusting the batch size or something along those lines

[rytaft]

Maybe you're thinking of #40850? I didn't end up merging that PR since there were some performance issues identified. Seems like we should just switch to vectorized to fix this.

[RaduBerinde]

I see. Let's use the existing #41203 to track this.