Use dask distributed scatter/persist for rayleigh correction

[djhoese]

I've been playing around with dask-distributed for using satpy/dask on a cluster. While learning more about how to submit tasks/data to workers I learned about the idea of sending data to the workers that stays at the workers for future calculations. This sounds like it could be a good idea for rayleigh correction where pyspectral/satpy pushes the LUTs to the workers and then as rayleigh correction tasks are submitted for processing they would be sent to workers that had the LUTs already in memory. I'm not sure this is a perfect use case for this since LUTs, by their nature, require the entire LUT to be available (you can't chunk them and do a look up in only one chunk).

I just wanted to write this down for future consideration.