Interesting MPI issue / Import health

[fdmalone]

In #230 I moved the from mpi4py import MPI imports to the MPIHandler class as importing MPI seems to call MPI_Initialize under the hood. This is problematic if subsequently one calls a subprocess which launches via mpirun / mpiexec (which is the case with the shciscf module.) Because mpi is already initialized mpirun does nothing and shciscf doesn't run. The hacky workaround was for me to delay the import MPI calls in mpi_handler, which were indirectly imported in the from_shci.py module.

There are two problems here, shci should probably live outside of the main repo (#229), and more broadly we should analyze imports as there is no real reason why mpi_handler imports should be making their way to from_shci which has no real dependency on our internal MPI calls (it's there indirectly through trial wavefunction or similar).

Just opening this as a reminder to myself, need to check if just using import mpi4py ... mpi4py.MPI is possible.