Make torch forecasting model loadable from BufferedReader

[MarcBresson]

Is your feature request related to a current problem? Please describe.

To load a torch forecaster, we need to provide a path. However, when the model is stored on the cloud (in a s3 for instance), it implies to first copy the file on the local storage to then give the appropriate path.

Describe proposed solution

The solution to that would be to add two arguments to https://github.com/unit8co/darts/blob/master/darts/models/forecasting/torch_forecasting_model.py#L1708 that are buffers (typically obtained with open("filename.pt", mode="rb") as in here). The first one would be the buffer to the base TorchForecastingModel, and the second one a buffer to the PyTorch LightningModule

Describe potential alternatives

None

Additional context

None