Auto label tree

[dlangenk]

When the auto-label tool is available (is there an issue for that?) we could implement a tool to automatically generate a label tree with the found species. This is of course that we can also do the object detection automatically.

1. You create a new volume/project
2. You click on a button or activate a checkmark while generating the volume which activates the auto label tree tool
3. In the background we do object detection (or random sampling) and use the auto-label tool to find relevant labels for this volume.
4. A new label tree will be added to the project with the found labels (potentially already with the found objects)

[mzur]

This is not how it will work (initially). To make the queries faster and the classification more accurate, the queries will be limited to the label trees of the project to which the new annotation belongs. So the users still have to specify what they are looking for beforehand.

Later we could decide to allow also other labels (of public label trees) that do not belong to the project, yet. But then I would just add the whole label tree to the project once the user decides to use a label of the tree. We should not create new label trees because these would duplicate the original labels and fragment everything even more (because a label cannot belong to multiple label trees).

Or we combine this with the idea of "virtual label trees" https://github.com/biigle/core/issues/554.