Introduce SPI for functions to change query structure

[beikov]

A few function emulations require some more sophisticated transformations like wrapping a query spec. I recently ran into a few functions, which on some databases, are only supported as window functions. To emulate the aggregate version, it is necessary to generate the window function version and push group by items into the partition by clause, and wrap the query spec to apply aggregation.

For completeness, I constructed an example of such a transformation.

select
	t.c1,
	rank(5) WITHIN GROUP (order by e1_0.theInt desc) 
from (values (1), (2)) t(c1) cross join (
	values (1, 5, '5'),
	(2, 6, '6'),
	(3, 7, '7'),
	(4, 13, '13'),
	(5, 5, '5')
) e1_0(id,theInt,theString)
group by t.c1

should be transformed to...

SELECT t2.c1,min(t2.c3) from(
	select
		t.c1,
		e1_0.theInt,
		rank() over (PARTITION BY t.c1 order by e1_0.theInt desc) 
	from (values (1), (2)) t(c1) cross join (
		values (1, 5, '5'),
		(2, 6, '6'),
		(3, 7, '7'),
		(4, 13, '13'),
		(5, 5, '5')
	) e1_0(id,theInt,theString)
) t2(c1,c2,c3)
WHERE t2.c2 = 5
group by t2.c1

I didn't think too much about this yet, but I think in org.hibernate.query.sqm.sql.BaseSqmToSqlAstConverter#visitFunction we should check if the SqmFunction implements an interface like QueryTransformer and apply that on the generated QuerySpec. I imagine the interface to look something like this:

interface QueryTransformer {
    QuerySpec transform(
			CteContainer cteContainer,
			QuerySpec querySpec,
			SqmToSqlAstConverter converter);
}

[beikov]

See #4764 for the implementation and a sample transformer for emulating ordered set aggregate functions when only window function variants are available.