RangeHeapJoin should consistently sort NULL values before non-NULL values while managing its heap.

[nicktobey]

Fixes dolthub/dolt#7260

This was ultimately caused by #1903. I didn't think it was possible for that issue to cause user-facing problems, but I was wrong. Because of that issue, RangeHeapJoins considered all NULL values in its children iterators to come after all non-NULL values. However, if the child node was an index, then the child iterator would order its rows with the NULL values first. This causes the RangeHeapIterator to mismanage the heap and skip rows that should have been in the results.

I updated the range heap code to manually check for NULL values when manipulating the heap. I also updated the plan tests to include NULL values in the test tables, which should now catch this issue.

[max-hoffman]

LGTM, our compare functions reversing null order seems wrong, I think I've had to do this same change in other places. If anything seems like auto-increment should be special cased instead at some point

[nicktobey]

Our compare function is absolutely wrong, but at the time the work to fix it (and update anything that was accidentally relying on the old behavior) felt not worth it. But if we're going to run into more issues like this, we should absolutely fix it.