[Patch, Fortran] PR fortran/35723: Some fixes to restricted-expression handling

[Dominique Dhumieres]

Daniel,

I know than generating errors is a subtle process. In the example I have sent, I think it
is important to generate the errors about "Expression at (1) must be of INTEGER type, found REAL"
because fixing them in the first place fixes also the other errors comming from the
bad declaration dimension(n,d).

Now I am not familiar enough with the gfortran code to give any advice on how the error
handling should be implemented.

BTW the original pr36192 gives an ICE after the error messages.

Dominique