Updates to fix nvhpc compiler issues

[mnlevy1981]

To run the standalone tests on derecho, we want to use nvhpc/25.9 (this includes a bug fix we reported where nvhpc/25.7 was not handling associate statements correctly).

The compiler also rightly complained about an instance of a missing kind number when using a real variable (although 1.0 is exactly representable so there shouldn't be any difference between 1.0 and 1.0_r8).

Thirdly, I replaced a couple of where statements with the do-loop if-statement combination to avoid dividing by zero. It is unfortunate that the compiler doesn't handle this very well, but it is not a very invasive change.

[mnlevy1981]

Note: I am trying to create a small example problem highlighting the issue with where statements in the nvhpc compiler, but my simple example doesn't create the same error I was seeing in the MARBL test suite. I expected

program where_statements

  real, dimension(3,3) :: old_vals, new_vals
  integer :: i

  old_vals(:,:) = 2.0
  old_vals(2,:) = 0.0

  do i=1,3
    where (old_vals(i,:) > 0.0)
      new_vals(i,:) = 1.0 / old_vals(i,:)
    else where
      new_vals(i,:) = 0.0
    end where
    print*, new_vals(i,:)
  end do


end program where_statements

to fail, but even with the same -O2 -Mfree compiler flags we use to build MARBL it runs fine. I don't think it's worth spending the time to further investigate / understand why I needed to modify compute_autotroph_nutrient_uptake() in this PR, although I can think of a few debug statements to include that might help highlight the issue.

[klindsay28]

LGTM