Quantity wrapped Dask Arrays need deterministic hashing

[rpmanser]

Based on xarray/#4208 (comment), Pint should implement deterministic hashing for Quantity wrapped Dask Arrays. This is a recommended step in implementing the Dask collection interface that I skipped in #1129.

This can be done in one of two ways: either implementing __dask_tokenize__ in the Quantity class, or registering a function with dask.base.normalize_token. The return value is the same. I'm wondering if there is a preference, since it seems that Pint tokenizes unit operations and I don't know if this would interfere with that. If there is no interference, I think __dask_tokenize__ would be better since it is cleaner.

I can follow up with a PR following any discussion.

[jthielen]

Given no other feedback, I'd say go ahead and try it with __dask_tokenize__. I agree with you that it is cleaner.

[rpmanser]

I'll put in a PR early next week