[dev] add torchvision detector functionality

[deruyter92]

Motivation:
In the existing code, the TorchvisionDetectorAdaptor is excluded for regular use (e.g. missing from the model registry). This means that pretrained torchvision detectors cannot be loaded automatically, requiring workarounds such as downloading the weights manually, and cleaning up the state dicts. An exemplary case is when working with the superanimal_humanbody_rtmpose_x from the modelzoo. This model requires a pretrained torchvision detector as it does not have any saved detector weights, and can therefore not be used directly in combination with the export functionality.

This PR aims to address this issue by including the TorchVisionDetectorAdaptor as a regular detector model, and updating the runner accordingly. The modelzoo export is also updated to enable export of the superanimal_humanbody_rtmpose_x

Changes:

• The TorchVisionDetectorAdaptor is registered as detector model
• The runner is adapted to load a model either when model weights are provided OR when config specifies pretrained=true
• A specific export functionality is added for the superanimal_humanbody_rtmpose_x, by including a hardcoded config (hacky..) .

    """Example usage"""	
model_name = "rtmpose_x"
super_animal = "superanimal_humanbody"
detector_name = "fasterrcnn_mobilenet_v3_large_fpn"

export_modelzoo_model(
    export_path= <my_model_path>,
    super_animal=super_animal,
    model_name=model_name,
    detector_name=detector_name ,
    )

[C-Achard]

Nice!
Is this ready to test, or how can I help ?
Otherwise what do you think a sensible way of exposing/documenting any built-in configs would be ? I think it's fine if we clearly centralize/document it.