Improve dataset selection in opener dialog

[hanslovsky]

Currently, datasets are always auto-discovered and then a list of choices is presented to the user, but

1. discovery can be slow for containers that contain many datasets
2. results are presented only after discover --- users have to wait!
3. long list of datasets can be annoying to search!

We should probably:

1. Add option to type in dataset (similar to n5 container with a menu button)
2. Present incomplete list with indicator that search is still going on
3. cache list of datasets (refresh cache on user request only)

This will greatly improve user experience when opening (multiple) datasets from container holding many datasets, e.g. on nrs

[hanslovsky]

Instead of discovering all datasets, it would probably be best to provide a file-browser like user experience:

1. auto-discover and present all the groups/datasets at the top level of the container
2. as users navigate through the container tree, auto-discover and present all datasets at the current level of the sub-tree
3. potentially cache (sub-)trees during navigation to already present choices while auto-discovery runs

This has the benefit of fewer options at a single time that the user has to pick from and selection will become easier from a less cluttered layout. fuzzy matching/ranking can still be useful at the tree level (or it will start discovery for the entire subtree).

[hanslovsky]

The JavaFX TreeView could be the right class for this.

[hanslovsky]

The TreeItem doc even has an example for a file browser that we could adapt.