fix: SortExec TopK OOM

[nuno-faria]

Which issue does this PR close?

• Closes SortExec with TopK causes OOM #17597.

Rationale for this change

The original code was referencing a record batch for each line in the limit, which was causing the arrow_select::interleave_record_batch to allocate vectors with billions of elements. For example, if we had 32 batches and LIMIT 50000, those 32 batches would be referenced 50k times. The new version uses only one reference per existing batch, limiting the memory used as well as improving the execution time:

> select *
from lineitem
order by l_comment;
-- Elapsed 1.804 seconds.

-- would OOM before
> select *
from lineitem
order by l_comment
limit 50000;
-- Elapsed 0.625 seconds.

What changes are included in this PR?

• Updated the way the record batches and indices are created in topk/mod.rs.

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

[adriangb]

Very exciting, amazing work!!