Is there any reason not to merge MemGroup and MemDiskGroup?

[nritsche]

@kiyo-masui after you had good arguments against the last proposal of changing the design, I'm hoping to get your opinion once again.

According to @jrs65 there's no good reason anymore to keep MemGroup and MemDiskGroup two separate classes.

[kiyo-masui]

I think it might be doable, but with significant labor. MemGroup exposes the same API as an HDF5 Group, but with the underlying storage in a tree of dictionaries. MemDiskGroup exposes the same API again, but is agnostic to whether the underlying storage is a MemGroup or an HDF5 Group (and then adds a bunch of functionality on top). I think it should work to merge the memory-specific code right into MemDiskGroup though.

I don't think there are compatibility reasons not to do away with MemGroup. We can provide a stub (MemGroup = MemDiskGroup) for compatibility, and they provide the same API.