dask: Data collapse methods and functions

[davidhassell]

Another hefty PR! Hopefully self explanatory, but these notes might be useful:

• Added Data.square and Data.sqrt because they were useful.
• Made some changes to Data.stats and Data.mean_of_upper_decile before realising that they can't be tested yet. So these will be dealt with in separate PRs. Feel free to ignore these methods.
• Needs dask v2022.03.0 which should be out later today :)
• Tidied up (i.e. removed) some redundant MPI stuff.

[davidhassell]

Hang on - forgot a test ....

[davidhassell]

OK - good to go.

[davidhassell]

Another note:

• I got rid of Data._map_blocks because a) it wasn't used very often, and b) to use it you really have to check what it does internally, and since it's only ~3 lines it made sense to me to in-line it.

[sadielbartholomew]

I'm mid-way through reviewing and everything seems good so far. Thought I'd raise some typos in the meantime, and if you can address the conflict resolutions that would help me to ensure I am reviewing the up-to-date code. Thanks.

[sadielbartholomew]

Overall this is fantastic. The updated code is well-structured and even copes sensibly across all of the statistics (bar those two methods you say are out-of-scope for this PR) for the slightly evil case of a fully-masked array. Great commitment to thorough documentation, too.

I've posed a few questions on minor aspects and suggested fixes for inevitable typos, one of which is in the functional code so is quite crucial to update. Thanks.