[BUG] Dropping NaNs on fit in all of the models

[alex-hse-repository]

🐛 Bug Report

We drop NaNs in fit here and here, however some models like Catboost are able to work with NaNs.

Expected behavior

We need to decide which behaviour is expected here:

1. Drop NaNs as we do now
2. Drop NaNs inside the Adapter where it is necessary
3. Do not drop NaNs

How To Reproduce

Environment

No response

Additional context

No response

Checklist

• [X] Bug appears at the latest library version

[alex-hse-repository]

Waiting for tests