[Experiment] Speedup map build

[Dandandan]

Which issue does this PR close?

• Closes Speed up hash join build phase [experiment] #18376

Rationale for this change

Performance improvements

What changes are included in this PR?

• Sort the items by hash (and offset) to be able to deduplicate hashes (this introduces some overhead but the hope is this pays off during inserting to the table)
• Use insert_unique (https://docs.rs/hashbrown/latest/hashbrown/struct.HashTable.html#method.insert_unique) rather than https://docs.rs/hashbrown/latest/hashbrown/struct.HashTable.html#method.entry for the first entry, which should be quite a bit faster by not having to search for existing items
• Keep on using the previous entry for duplicated elements (saving calls to entry for each duplicate)

Are these changes tested?

Yes. existing tests

Are there any user-facing changes?