fms2_io: Add collective writes

[J-Lentz]

Description
Adds parallel writes to fms2_io for 2D, 3D, 4D, and 5D cases with a new unit test.

Fixes #1715
How Has This Been Tested?
Builds with ifx 2025.2.0 on the AMD box and the test passes.

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published in downstream modules
• New check tests, if applicable, are included
• make distcheck passes

[uramirez8707]

As we discuss on Monday, if using collective mpi netcdf the code is going to put out a NOTE stating that the io_domain will be ignored and only write out 1 file.

These are some other changes that are needed:
uramirez8707@6999342 This is so that the dimensions are registered with the correct size (the compute domain rather the io compute domain)

uramirez8707@5908a72 This is so that the code doesn't write the domain decomposition attribute since it is not needed since there is no combining

[uramirez8707]

Please include this commit so that there is a function that we can use to determine if a file is using collective mpi netcdf:
uramirez8707@c0f668b

I found this useful when implementing this in the diag manager

[uramirez8707]

Rather than initializing e to 1, do

  e(1:2) = shape(vdata)

And for the 3D case,

e(1:2) = shape(vdata)
e(3) = 1 ! unlimited dimension

[bensonr]

Is there a requirement for the last index to be the unlimited one? Will this have any effects on future efforts to implement generalized indices?

[J-Lentz]

I think we've been assuming so far in the generalized indices work (at least, I did in the test that I worked on) that the unlimited dimension would always be last.

It's not clear to me if this is a hard requirement. It seems like it might not be with NetCDF 4, but the NetCDF documentation also says (https://docs.unidata.ucar.edu/netcdf-c/4.9.3/unlimited_dims.html):

For programmers, the unlimited dimension will correspond with the slowest-varying dimension. In C this is the first dimension of an array, in Fortran, the last.

[uramirez8707]

Very minor documentation changes, but looks good otherwise.

Nice work on the tests!