Optimize subquery transformation for SIZE() select clauses

[Mobe91]

We could use the inverse mapping for creating the COUNT() subqueries. This way we could save a join.

[beikov]

I would like to recap what the intention of this issue was since it's very old now.

Right now, we transform SIZE functions into subqueries that do COUNT. When having

SELECT SIZE(a.collection) FROM A a

it get's transformed to

SELECT (SELECT COUNT(y) FROM A x LEFT JOIN x.collection y WHERE x.id = a.id) FROM A a

although it could be transformed into

SELECT (SELECT COUNT(y) FROM B y WHERE y.a.id = a.id) FROM A a

if we know the inverse relation.

According to the JPQL BNF we could also transform it into the following equivalent query

SELECT (SELECT COUNT(y) FROM a.collection y) FROM A a

which would be even easier.

[beikov]

Fixed