Account for Fortran assumed shape

[joewallwork]

Closes #2820.

As documented in the issue, I wasn't able to build the allocatableArraySimple example locally. It seemed that Enzyme didn't fully account for assumed shape arrays in Fortran. The small patch in this PR addresses this and I can confirm I can now forward and reverse mode differentiate through things like

  subroutine normalise(x, y)
    real, dimension(:), intent(in) :: x
    real, dimension(:), intent(out) :: y
    y(:) = x / sum(x)
  end subroutine normalise

which is really nice.

If I understand correctly, the error message I was hitting was slightly misleading so I propose an alternative phrasing.

[joewallwork]

This all works well on it's own. However, I've just discovered that it doesn't work in conjunction with #2809. I tested this branch merged with that one on cases with assumed shape arrays and they compile but produce segfaults. This branch also produces a warning like

warning: binary operator is integer and assumed constant:   %19 = mul nsw i64 %11, %15

for each assumed shape array encountered, which isn't ideal. For a full fix, we should address #2822.