Having aggregate alias bug

[max-hoffman]

fixes: dolthub/dolt#7082

[nicktobey]

I added some additional tests to the branch that I wrote while trying to convince myself this is correct.

The fundamental issue is that aggregate expressions should always prefer binding names to columns in the FROM expression over aliases in the SELECT expression. But if there isn't a match in the underlying tables, we're allowed to match to an alias. This can be correct behavior if the alias is uniquely determined by the aggregated column.

This can be seen in one of the examples I added:

select y as z from xy group by (y) having AVG(z) > 0;

z does not bind to a column in xy, so it binds to the alias. The alias is uniquely determined by the aggregate column, so the behavior is well-defined.

With this PR, that statement is an error. But it should be allowed.

[nicktobey]

On the other hand, it looks like the added test doesn't work before this PR either, so this PR may still be making things strictly more correct.

[max-hoffman]

This one is a good test, but I'd separate it as an additive change to what I have so far

select y as z from xy group by (y) having AVG(z) > 0;

We can only express one group by per scope, so the z is going to have to be variable replaced and normalized into the format of other aggregations.