Fix current issues with forecasting

[americast]

This PR aims to solve the following issues:

• Throwing error when non-numeric characters are in the data (partially fixes Enhance forecasting with more features #1243)
• Math domain error with statsforecast.
• Fix GPU support for neuralforecast.
• Neuralforecast support for directly using batched data.
• Auto frequency determination ( Better frequency inference and message for timeseries forecasting  #1279).

Will create separate PRs for the last two points.

[xzdandy]

Hi @americast

1. We can have separate PRs instead of one PR for everything.
2. For non-numeric characters are in the data, it is not only a problem for forecasting, for example, classification/predictions may also suffer this data cleaning problem. I think the alternative approach is to provide a REGEX function for generic uses instead of forecast only.

Let me know what you think. You don't need to worry about the REGEX function, and can look into forecast-specific issue more.

[americast]

Sure. Will create separate PRs for pt 4 & 5. Keeping this single PR for the three short changes as mentioned in pts 1, 2 and 3.

[americast]

@xzdandy Resolved all the concerns

[xzdandy]

I am fixing the Doc.

[xzdandy]

I will update the tutorial notebook to showcase these two features.

[americast]

I will update the tutorial notebook to showcase these two features.

Thanks a lot @xzdandy! Let me know once you are done with that add I will add some more comparisons to it.