Introduction of aggregation datasets

[davidhassell]

Fixes #319

Moves CFA functionality from cf-python to cfdm

I think fairly self explanatory (let me know if otherwise :)), but some points of note:

• CFA API is updated a bit to reflect that which is (currently) going through discussion over at the CF conventions
• read.py is refactored as a class so that upstream applications (e.g. cf-python) can avoid large amounts of docstring and code duplication
• The inheritance structure of Array objects has been changed for similar reasons. This includes moving the old FileArrayMixin to a the abstract FileArray

[sadielbartholomew]

Looks all good, with sufficient testing and docs updates. I've raised only minor comments (mostly typos). Once you've considered these, please feel free to merge.

[sadielbartholomew]

Thanks for responding to the minor feedback items. Everything seems to be covered now. I've done a final sanity check and all is good, except for a few new typos introduced, as raised in-line. Please blitz those and then we can merge.

[davidhassell]

Thanks for all the improvements, Sadie! merging now ...