Bug fix for GPU-lized PUMAS codes

[sjsprecious]

Recently we realized some bugs in our GPU-lized PUMAS codes and we would like to fix them in this PR.

The detailed changes of this PR include:

• Fix a bug in the Sedimentation subroutine that will lead to wrong GPU results.
• Replace the #ifdef variable by a Fortran parameter.
• Switch the acc declare copyin to acc declare create and acc update device, which is BFB for GPU but requires fewer data copy.

These changes are BFB for CPU results.

Test suite: [aux_cam] Test machine: Cheyenne Test results: [BFB]

[sjsprecious]

Fine as long as it doesn't change answers. Seems like this bug might have affected the CPU code for sedimentation, or were we not passing i,k vectors?

Thanks @andrewgettelman . The CPU code for sedimentation will be changed too, but it is BFB.