Zarr write + Zarr groups

[davidhassell]

Fixes #354 and #355

Orientation

cfdm/data/aggregatedarray.py

• Change to allow URIs to be Python strngs or scalar numpy string types.

cfdm/data/data.py

• Include Zarr dataset sharding methods
• Include Zarr fragment class
• Move the test for an "empty" slice to the parent construct (in cfdm/mixin/propertiesdata.py), because sometime we want to allow slices on abstract data (just like numpy does)

cfdm/data/fragment/fragmentfilearray.py

• Include Zarr fragment class

cfdm/data/fragment/fragmentzarrarray.py

• New file defining a Zarr fragment (very similary to cfdm/data/fragment/fragmentnetcdf4array.py

cfdm/data/netcdfindexer.py

• Account for zarr using the new numpy T data type

cfdm/mixin/netcdf.py

• New mixin class for dataset shards

cfdm/read_write/abstract/abstractio.py

• Rename "file" to "dataset"

cfdm/read_write/netcdf/flatten/flatten.py

• Rename "file" to "dataset"
• Use match clauses to switch between dataset format APIs
• Changes to allow Zarr datasets to be flattened
  • Zarr datasets to not have well-defined Dimensions, so finding
    which dimensions belong to which groups for Zarr datasets is more
    involved, and needs configuring with the new
    group_dimension_search keyword.

cfdm/read_write/netcdf/netcdfread.py

• Rename "file" to "dataset"
• Use match clauses to switch between dataset format APIs
• Handle Zarr shards
• Performance improvements in _cache_data_elements, largely aimed at
  reducing the number of requests to disk

cfdm/read_write/netcdf/netcdfwrite.py

• Rename "file" to "dataset"
• Use match clauses to switch between dataset format APIs
• Handle Zarr shards

cfdm/read_write/netcdf/zarr.py

• Include a reference variable in ZarrDimension

cfdm/read_write/read.py

• New keywords store_dataset_shards and group_dimension_search

cfdm/read_write/write.py

• Rename "file" to "dataset"
• New keyword dataset_shards

setup.py

• zarr import optional