PCOMM: Rename Precv_init/PSend_init to MPIX for 5.0.0

[hppritcha]

Related to #11967

The MPI Forum decided to change the API for partitioned communication subroutines MPI_Psend_init and MPI_Precv_init as part of MPI 4.1. This has generated quite a bit of dicussion, see mpi-forum/mpi-issues#765.

Rather than release a 5.0.0 with a set of functions which will be redefined if the current MPI 4.1-rc is ratified and we want OMPI to be 4.1 compliant, just rename the existing MPI_Psend_init and MPI_Precv_init to MPIX_ prefixed.

This commit also fixes a bug with the Fortran interfaces. They weren't compliant with the MPI 4.0 definitions in any case.

[bosilca]

Kudos to whoever did the implementation of these two functions, because she did not followed the API from the standard but instead named these functions correctly (both MPI_Precv_init and MPI_Psend_init use ints for count). Thus, there is no need to change the type of the count argument to MPI_Count in this patch because:

1. this is the sensible naming scheme and we can hope that at the end the standardization will legitimized this API
2. we are in the MPIX namespace and we can have the API we want and having a sensible naming scheme is important.

[bosilca]

Why changing the count from int to MPI_Count ? To follow the broken MPI API ?

[hppritcha]

okay MPI forum decided today to back out the proposed changes to the 'c' MPI_PSEND_INIT/MPI_PRECV_INIT to MPI 4.1.
So the thing now is to decide whether we want to "fix" the Fortran bindings - which are not compliant with the MPI 4.0 standard.

[jsquyres]

Given that the problematic proposal has been withdrawn from consideration for MPI-4.1 -- and API/ABI will not be broken in MPI-4.1 (or later) for these 2 functions -- we should no longer rename these functions to MPIX.

Instead, we should fix the Fortran bindings to be what is in MPI-4.0.