Integrate neural process code into experiment/fitting workflow.

[noseworm]

The experiment workflow allows us to manage all the training artifacts across training/fitting/testing phases. This issue keeps track of everything that needs to be done to make the neural process code compatible with this infrastructure.

Training Phase

• Integrate train_grasp_np.py into run_grasping_experiments.py
• Save best validation model. Separately save the Decoder wrapped in an Ensemble then LatentEnsemble object (for now, use 1 ensemble model, but can consider training an ensemble of decoders in the future).
• Ensure that the LatentEnsemble/ Ensemble objects work: should be able to do a forward pass using a particle distribution for known objects.

Fitting Phase

• The fitting phase code expects a LatentEnsemble object. The previous steps should ensure this works.
• We will need to integrate the dataset preprocessing code (create_gnp_data.py) into the sampling script to make sure the decoder gets the expected input.