Add the ability to set chunksize when making the dask array

[Cadair]

This PR got a little 🐂 shaving, I removed the use of ExternalArrayReference, however, its primary purpose is to provide the primatives to allow the optional specification of a chunksize when making a dask array from each loader referenced by an asdf.

This should be all the changes needed to test the performance of loading a chunked array, and then changes to the asdf maker can use them later.

This also cleans up the implementation of the Loader class by making the abstract method the __getitem__ method, this is needed to access .section and read only a subset of the file.

TODO:

• Write a test jig which loads an asdf file and then retrospectively injects a chunksize and benchmark it.

[codecov]

Codecov Report

Merging #232 (6abebee) into main (feb7bb2) will increase coverage by 0.01%.
The diff coverage is 95.40%.

@@            Coverage Diff             @@
##             main     #232      +/-   ##
==========================================
+ Coverage   97.18%   97.20%   +0.01%     
==========================================
  Files          37       37              
  Lines        2095     2108      +13     
==========================================
+ Hits         2036     2049      +13     
  Misses         59       59              

Impacted Files	Coverage Δ
dkist/io/asdf/converters/file_manager.py	92.59% <66.66%> (-3.41%)	⬇️
dkist/io/dask_utils.py	72.22% <83.33%> (+1.63%)	⬆️
dkist/io/loaders.py	96.00% <90.90%> (-4.00%)	⬇️
dkist/dataset/dataset.py	92.04% <100.00%> (-0.27%)	⬇️
dkist/io/asdf/entry_points.py	100.00% <100.00%> (ø)
dkist/io/asdf/tests/test_dataset.py	100.00% <100.00%> (ø)
dkist/io/asdf/tests/test_tiled_dataset.py	100.00% <100.00%> (ø)
dkist/io/file_manager.py	98.90% <100.00%> (+1.58%)	⬆️

[SolarDrew]

LGTM 👍