Introduction of Dask for all data manipulations

[davidhassell]

It has become necessary for Dask to be used in cfdm for data manipulations, so that we can support aggregation datasets (previously known as CFA datasets), which are going to appear in CF-1.13 (see cf-convention/cf-conventions#508).

Note that we have to currently pin Dask to <= 2024.7.1, as a consequence of this Dask issue: "Slicing problem that can occur when shuffle algorthm for 'take' is applied" (dask/dask#11669).