When parallelizing existing numpy codes it would be convenient to support operations like -=, +=, etc.. This won't have any performance benefits, but would be somewhat convenient.

Typically we do this by doing the normal operation and then overwriting .dask, .name, etc..