fix ignoring [MaybeNull] during type inference

[TomatorCZ]

The fix regards the issue

Problem
The parameter field TypeWithAnnotations doesn't reflect [MaybeNull] attribute during type inference of method parameters.

Result
Different behaviour of mentioned examples in the issue. The first example behaves like this method TestA<T>(IOperation<T> operation, out T result) resulting in T = string instead of TestA<T>(IOperation<T> operation, out T? result).

[jcouv]

Nullable attributes are not meant to contribute to method type inference. That part was by-design from what I recall.
The linked issue (#50782) mentions an inconsistency between out string? discard1 and out string? _. That does seems like a bug.

[RikkiGibson]

I'm not familiar with the decision re: nullability attributes and type inference. It does seem to me that in the out T? param case (TestB in SharpLab), we might not be contributing nullable bounds correctly.

Just because an out string? argument contributes string for T, it should not be an exact bound, so it should still be possible to choose string? as the type argument based on new StringOperation() contributing string? (maybe an exact bound, since IOperation<T> here is invariant?)

Would be interested to discuss further offline.

[TomatorCZ]

I close this PR based on offline discussion with @RikkiGibson .

MaybeNull means different thing compared with ? notation.

Part of the fix continues as a separated PR regarding different behaviour of discard expression.