Fix AND/OR negation logic in hql

[csharper2010]

AND/OR below NOT must use result of subsequent NegateNode calls to create a new result node instead of partially mutating the existing node to consider functional signature of the NegateNode method

Fixes #3327

[bahusoid]

Hibernate similar fix for reference: hibernate/hibernate-orm@aac7e3e

[fredericDelaporte]

For inclusion in 5.4.x as asked in its corresponding issue, #3327, it would need to target the 5.4.x branch and being based on it.

The test case needs to be fixed for some databases too. (I have not checked what is the trouble. It could be that they cannot support what the test does. In such case the test needs likely to do some feature support checks as done in some other tests. It needs investigation.)

Doing this will not guarantee we will merge it in 5.4.x: there is always a risk of introducing some new regression with each change. Ideally 5.4.x releases are meant to stabilize (fix regressions) the 5.4 minor version. Additional fixes added into it may delay its stabilization. And also, adding new fixes in it instead of vNext (5.5) is not an incentive to finish 5.5.

Still we do it sometimes.

[csharper2010]

The test case needs to be fixed for some databases too. (I have not checked what is the trouble. It could be that they cannot support what the test does. In such case the test needs likely to do some feature support checks as done in some other tests. It needs investigation.)

There seems to pe a foreign key problem on setup of the entities or the db structure is not generated as expected. I will have a look next week.