Support selected new `numpy`-consistent `Data` methods

After or during the LAMA to Dask conversion (#182) we should review the Data methods we support, especially because LAMA didn't yet support binary operations in a performant way, but Dask enables those, and particularly also for data operations like mathematical functions where we don't seem to have very good coverage relative to the possible methods supported by NumPy (see ours vs. numpys).

The following can be used as a list of methods to add, or to prioritise adding; please add to the list as appropriate, particularly with user requests or where a method seems particularly useful in the context of cf to add (try to include some contextual notes):

• arctan2: already largely coded up but commented out as it is a binary operation and therefore was not possible to do in a performant way with the current LAMA (see #38);
• sqrt: user request by @bnlawrence (I pointed out f**0.5 works for this, etc., but the method form was preferred and requested);
• cbrt: why not add if we are adding the above anyway, where all other more niche roots can be covered by the ** operator;
  ...

Start by reviewing the LAMA-to-Dask conversion in #182 and the existing commented-out arctan2 implementation, then compare cf.Data’s mathematical-functions documentation with NumPy’s routines. Prioritise the requested arctan2, sqrt, and cbrt methods, and consider the work done when the selected methods are supported consistently with NumPy and documented.