Adds Region Of Interest (ROI)

Optionally, a ROI can now be configured. Only objects within the ROI are included in the state count. I am using this to detect when there is a car outside my house, and ignore cars not outside my house.

Closes multiple small issues

This release closes multiple small issues, but also gets a significant version number update as there are a couple of breaking changes.

1. The config variable target becomes targets, as it is a list
2. The file_saved events are removed, since these were always buggy and the recommended approach is to use folder_watcher to detect when new files are saved.

I have a few more impovements planned, but should not be any more breaking changes.

Rename events to use prefix deepstack

Rename events to use prefix deepstack to be in line with new HA best practice. Also hotfix #102

Supports multiple targets

Thanks for @shbatm you can now provide multiple targets in the config like this:

target: person  # original, still works
# New:
target:
  - person
  - car
  - truck

Will match all possible targets above the confidence level and return the total count as the state (also some minor lint and formatting cleanup).

Fix import for HA 0.103

Fixes an import error introduced by HA 0.103

Fix race condition and correct timestamps

Fixes a (potential) race condition where a file could be called in an automation before it was updated. Also addresses #90

Makes save_timestamped_file opt in

This minor release adds a config variable save_timestamped_file which makes the saving of timestamped files opt in. Additionally centroid info is dropped

Use entity name in saved files

Couple of small changes:

• The entity name is now used in the file name of saved .. _latest.jpg images
• The default scan_interval is 365 days, so in practice it will not scan unless you call the scan service. Alternatively you can configure a scan_interval for whatever period you like
• Some under-the-hood refactoring
• Bounding boxes are now red (rather than yellow)

Publish box data

The bounding box and object centroid data are now published in the image_processing.object_detected event. This completes a transition to an approach where the UI is for typical users, and events are for power users.

<Event image_processing.object_detected[L]: entity_id=image_processing.deepstack_object, object=dog, confidence=99.9, box=[0.559, 0.674, 0.842, 0.83], centroid=[0.752, 0.701]>

Additionally the target is not listed in the attributes as it is the unit_of_measurement. This streamlines the UI:

Tidy attributes and events

Tidies the attributes and events. Overall there is less noise in the UI, and less duplication of information. For confidences of detections you now need to see the events.

Note I accidentally pushed to master, so no PR to review, oops