[Feature] Implementing MPI_IN_PLACE for reductions

[PhilipVinc]

Hi,

I would like to implement functionality to perform reduction with MPI_IN_PLACE, so that the output buffer is the same as the input buffer.
I would gladly contribute this back to the repository, if someone could guide me towards implementing this correctly.

For Allreduce I'm thinking about adding a method to mpi-base.jl with this signature

Allreduce!(sendbuf::MPIBuffertype{T}, op::Union{Op,Function}, comm::Comm) where T = 
    Allreduce!(Ptr{Cvoid}(1), sendbuf, length(sendbuf), op, comm)

something similar can be done for Reduce.

Also, Ptr{Cvoid}(1) corresponds to the definition of MPI_IN_PLACE=(void*)1 taken from mpi.h.
I am uncertain on what is the correct way to include it so that it becomes available as MPI.MPI_IN_PLACE.

[PhilipVinc]

I implemented the above into philipvinc/in_place branch at https://github.com/PhilipVinc/MPI.jl/tree/philipvinc/in_place

At the moment I have added support for

• Allreduce, Allreducev
• Allgather

If a maintainer confirms that you are interested in this, and helps me with a list of methods to support, I would gladly add those, along with tests, and submit a PR.

[PhilipVinc]

Closed by #232