[ENH] Request for model refit

[jobs-git]

Is your feature request related to a problem? Please describe.
Refit an existing model to enable improving the model by updating parameter based on fresh data.

This feature would be useful for:

• batching really huge data, something that cant fit to vram
• continuing the training from an interupted process. So we need to also have the ability to model.save in between epochs.

Describe the solution you'd like
So we need a model load, then followed by refit:

model = nbeats.load(path)
model.refit(y)

the loaded model should have all the information it has including epoch, fh and so on, so we can ensure the model is trained with new data on the same condition and parameters, so it would appear like you just have a new batch coming in, like it is trained in one go

Describe alternatives you've considered
n/a

Additional context
n/a