multi channel segmentation support

[saeliddp]

Context

See #374 . In order to implement support for multi-channel segmentations, I found it easiest to do some light refactoring to use slots/signals and our task executor framework. This is definitely not a full refactor of training to align with curation, but it is a start.

Main Changes

1. channel_extraction.py: add an optional arg to the function that gets an image path from a provided csv.
2. training_model.py: add state related to segmentation channels, remove unused state and associated functions
3. training_service.py: implement async directory data extraction using the task executor pattern
4. image_selection_widget.py: add UI elements for selecting segmentation channels + handlers for what happens when channels are set/selected
5. cyto_overrides_manager.py: add overrides for segmentation channels

Testing

Added unit tests for image_selection_widget.py and training_model.py. Fixed any failing unit tests. Edit: I have now manually verified that we can run through curation/training with channel overrides for segmentations.

Screenshots

[hughes036]

If this implements the plugin side of the contract with cyto-dl, i think we can merge. if the contract changes, we can make a followup pr.

[saeliddp]

@hughes036

If this implements the plugin side of the contract with cyto-dl, i think we can merge

It does, but if we merge without the corresponding changes in cyto-dl being released, training will fail