provide a way to access fedbiomed.common.models.Model of a TrainingPlan

[mvesin]

Several portions of code need to access the training plan's Model class, but this currently requires accessing the training plan's private variable, which is not clean;

self.training_plan()._model

This exists in TrainingPlanWorkflow.breakpoint, Experiment.run_once, ...

As a developer, I want to provide a cleaner way to access the Model class.

1. simple implementation (done in this issue)

Add a function to TrainingPlan with a name complicated and clearly different from TrainingPlan.model() to avoid unwanted access by user, eg TrainingPlan.get_model_wrapper_class()

2. more complete implementation (mentioned for future evolution, not implemented here)

Re-implement Model class as a subclass (decorator) of nn.Module, SGDClassifier, etc. to avoid duplication of objects and access methods

TorchModel should be a "transparent wrapper" (I think the pattern is often called a decorator). It should retain all of the functionalities of nn.Module, plus some added interface. This would allow researchers to use it transparently as they would use the torch nn.Module, but internally in Fed-BioMed we could also call the added functionality (e.g. get_weights) when needed.

[mvesin]

hi @sharkovsky I found that we also have such access from DPController class to optimizer._model

Should we also add a get_model_wrapper_classer to BaseOptimizer ?