MatrixMult bcast #168
Description
In #160 the following line https://github.com/astroC86/pylops-mpi/blob/51551867fb61651ad9d3d3fb1d22c522b86ec1a4/pylops_mpi/basicoperators/MatrixMult.py#L603 is implemented using buffer based MPI4py BCast; whilst testing with object based MPI4py bcast, @astroC86 reported some problems.
Whilst we ultimately want to implement all communications with object based MPI4py methods, right now we currently implement everything with object based MPI4py bcast; from one hand this is slower, from the other this is so far more portable to CuPy+MPI as it works also when CUDA-Aware MPI is not available. The ultimate solution, so far sketched in #167 will have all options in place and automatic dispatching based on a mixture of env variables and internal variables, so we need to also make sure that this code runs with object based MPI4py bcast.