refactor!: mpi hdf5

[drodarie]

Add MPILock to static methods to load hdf5 file.

Fix dbbs-lab/bsb-core#893
Needs dbbs-lab/bsb-core#902

Closes #16

[Helveg]

Neither this or the parent PR are fix PRs, they're a breaking change to the storage interface. Maybe this is a good time to include a dev doc in the public documentation that documents all the pieces of the Storage interface, and to see if there's other opportunities of stuff to rectify. I'd for sure make sure that all method signatures are sort of the same, and that all the methods are MPI aware (since that seemed to have been overlooked for some?) and have access to the communicator.

[Helveg]

is a breaking change, needs an extra moment to see if there's other problems

[drodarie]

I do not understand why we do not simply clear the chunks' group in the hdf5 file when we clear their content.
In Chunkloader.clear.

Is there a reason why they should not be removed for good?

    @handles_handles("a")
    def clear(self, chunks=None, handle=HANDLED):
        if chunks is None:
            chunks = self.get_loaded_chunks()
        for chunk in chunks:
            del handle[self.get_chunk_path(chunk)]

[Helveg]

Is there a reason why they should not be removed for good?

I don't recall, you can give it a go and see if anything breaks.