Implement in-sample and out-sample predictions in `nn` models without exceptions

[Mr-Geekman]

🚀 Feature Request

Currently we can't make with nn models

• in-sample predictions
• out-sample predictions there ts isn't going straight after the train and only on horizon steps in the future We are throwing exception because of limitation of implementation. However, it is possible and we can implement it.

Proposal

Add possibility to make

• in-sample predictions
• out-sample predictions with gap

Probably it will be easier to do after merging #776.

Test cases

1. Make sure current tests pass.
2. Pay attention to tests in test_inference, after making in-sample and out-sample with gap possible without the exception, we should probably move nn models to a new test case.

Additional context

This task waits for #787 to finish.