Rather than passing an optional `converter` to methods,
we now store staged files with a specific reader key.
The key relates to an entry-point (in group `jcache.readers`) of dynamically loaded reader.

Also, the `jupyter_executors` entry group has been changed to `jcache.executors`,
and `importlib-metadata` is used to load entry points.

It was felt that this is conceptually easier to understand,
i.e. it is a list of records for each notebook (& associated data) in the project,
rather than just a staging area for pre-executed notebooks.

For `jcache project` commands, and remove `--all` option,
in favour of separate `jcache project clear`

To write out a notebook merging the project file with its cached outputs.

The execution logic was also refactored, to reduce code duplication.

Note, artefact retrieval has been removed for now,
until the logic can be improved.

Co-authored-by: Chris Holdgraf <[email protected]>

click v8 has improved completion handling

So that we can make it more flexible, and allow for the myst-nb custom formats (with kwargs)

…to caches, move `--cache-path` to sub-levels