Optimization: Parallelize finalize state of aggregation

[anuchak]

Description

Currently, the finalizeInternal function of HashAggregate operator is performed in a single threaded manner.
For aggregations performed on large tables, the finalize becomes a significant bottleneck.

I'm running benchmarks on the MS MARCO dataset for FTS where we do aggregation for creating the index: https://trec-rag.github.io/annoucements/2024-corpus-finalization/

On a small segment partition (#00), the following query:

MATCH (b:ms_marco_test) WITH tokenize(b.segment) AS tk, OFFSET(ID(b)) AS id UNWIND tk AS t
RETURN STEM(t, 'porter'), id, count(*);

takes 134322.01ms to run and just the finalize part takes 84198 ms.

[acquamarin]

Duckdb implements the parallel aggregation algorithm described in the section4.4 of this paper: https://15721.courses.cs.cmu.edu/spring2016/papers/p743-leis.pdf.
We can consider follow the idea.