Introduce Experimental API - Future features

[mert-kurttutan]

Is your feature request related to a problem? Please describe.
Introduce experimental submodule to provide less stable experimental API

Describe the solution you'd like
Put the submodule right below the main module, .i.e. childe module of main module.

Describe alternatives you've considered
None

Screenshots / Text
Not Needed

Additional context
This will be useful for the prototyping some experimental features, see below

[mert-kurttutan]

Features to increase the flexibility of torchview:

1. Add an option to draw computational graph of methods other than forward
   For instance, we would want to have

   model_graph = draw_graph(model, input_data, model_method="preprocess")

   Issues:

   • record_tensor module is based only on __call__ method. We would need to adapt it based on the method provided.

2. A context manager to record computation into visual graph: This seems pretty self-explanatory. Create a context manager, e every (torch) operation inside it.

   An example usage:

   with draw_graph(model):
     out = model(in_tensor)
     pred = model.special_infer(in_tensor)

   And you get the computation graph for the entire computation inside the context manager.
   Issues:

   • record_tensor module is based only on call method. We would need to adapt it based on the method provided. This is because a variety of operations will be called inside the context manager. An ideal tool should capture this, otherwise, it is kind of pointless to introduces this feature
   • Register a model with context and manager and its respective methods.