Small issues to fix

[JonasFrey96]

• Within all extraction scripts make the configuration nice and add testing for it
• Rework the parsing of the dimension of the input features from DINO
• If mean kernel doesnt work in DINO remove all the code for it
• Implement the ConfidenceGenerator in AnomalLoss - replace the mean calculation
• In traversabilitzy estimator remove unused dependencies and remove all ros dependencies
• Publish the reprojection mask within the learning node in the imagefeat_callback only if debug is ture
• Check the implementation for the pervious nodes if this is needed
• Check if the loading of the different models works
• Remove in all files unnecessary default params
• Implement confidence generator together with the existing clipping in the loss
• Remove unnecessary python library dependencies
• Refactor the configuration - make the dataclass contain all variables that are also set via the ros parameters otherwise it is very hard to keep track.
• Remove the hierarchical learning folder the flatter structure improves readability
• Update in robot file reload_default_params to False

[RobinSchmid7]

Mean kernel does not seem to help a lot, seems to create more noise in the lower half of the image. Remove part of this code.

[RobinSchmid7]

• Add namespaces to topics

[RobinSchmid7]

• Implement correct stopping of ros nodes when running robot.launch