Invalid date parameter format with nullables and MappedAs

PR done.

The expression constant seen at HQL tree construction is still a DateTime? because a second partial expression tree evaluation occurs from NhRelinqQueryParser, after the parameters map build in NhLinqExpression, when its Translate method is called.
The parameters map build alters the expression by removing the non "evaluatable" MappedAs call, yielding instead its first parameter value (the DateTime in this case). This causes the wrapping convert to nullable to be "evaluatable", while it was not when the MappedAs was still there.

Then the second partial expression tree evaluation further alter the expression, transforming the DateTime constant to a DateTime?, causing it to be no more "parameterized".

All that is worsen by the fact that the expression Key has already been computed before this second evaluation, causing it to ignore the DateTime value since it is supposed to be "parameterized". This causes the additional parameter value caching bug.

The NhRelinqQueryParser partial expression tree evaluation seems unneeded, since it is used solely by NhLinqExpression, which has already done it within its constructor. It looks safe to me to remove it (and all tests passes, going to PR that). Unfortunately, this may be a possible breaking change for some external library, since NhRelinqQueryParser is public and could have other usages outside of NHibernate.

It seems the root of the trouble is the comparison of a "to be parameterized" DateTime to a nullable property: the resulting expression tree wrap the DateTime inside a convert causing it to be identified as a parameter candidate with a nullable DateTime? type.
But when we additionally call MappedAs on it, the wrapping convert can no more be evaluated by Re-Linq since MappedAs cannot be evaluated. This causes somehow (still have to identify where precisely) a parameter type mismatch: the ConstantToParameterMap ends up filled with a non nullable DateTime key (in ExpressionParameterVisitor.VisitMethodCallExpression) while the hql tree constant expression is still a nullable DateTime? (not determined yet from where it comes). This causes it to be not parameterized in HqlGeneratorExpressionTreeVisitor.VisitConstantExpression.

I wonder if we should strive for eliminating this mismatch, or if we should support it by making ConstantToParameterMap considers such values (one nullable, the other not, but having the same value indeed) equals.

Well, there is actually a parameter value caching bug, but its reproduction conditions are so closed to those of this bug (requires comparison to nullable property and usage of MappedAs) that it is very likely to have the same root cause.

So I have just added some additional test cases in this bug.

Test cases available here.

I intend to try fix this, but not today. I think I have found another more critical bug while elaborating those test cases, something similar to but with datetime parameters. I will try to sort this out first (and then issue another Jira if I end up confirming this another issue).