Add batch inference

[egoriyaa]

Hi!
As I correctly understand, if dataset contains a lot of time series and inference is doing on gpu, model can fail due to OOM.
May be it will be better to pass "batch" of time series to model.

[lostella]

@egoriyaa of course, you can chunk your inference dataset in batches, and pass them to the model one batch at a time. We don't do that in the package since it's something that users can easily do around the .predict call in the way that is most convenient for them.

Autogluon (see tutorial) does this out of the box, if you're interested: you pass in a dataset, and it will take care of loading it in batches, see these lines.