Learning-loop-node

Detector fix:

- Don't set log level of outbox logger to debug

Fix in Yolo Trainer Node:

- Prevent too many requests when downloading images from the loop

Fix:
- Improve SIO reconnection mechanism

Features

- Major improvements regarding the mechanism to (re)connect to the learning loop
- Improve openapi documentation

Detector features:
- New detection field `source`. Whenever the detect function is called (REST or SocketIO) it is now possible to provide the `source` of an image. This string should e.g. identify the robot in terms of its ID. The information is passed to the learning loop.
- The /about REST endpoint now contains the current state of the detectors `version_control` state ("follow_loop", "pause", ...)
- Less and configurable logging. The level of the loggings that were made when `detect` was called are now set to debug. However, there is now a REST endpoint that allows to enable/disable debug logging.

Detector/Trainer features:
- Configurable backdoor controls. Whether or not a node should use the REST backdoor controls can now be set via an environment variable (see README.md)

Detector dependencies:
- We removed the dependency to the fastapi_socketio lib (which is no longer maintained)

Trainer Nodes
Bug-fixes
- Prevent Division-by-Zero Error if all images are already downloaded