[OpenSTEF 4.0] Save rolling aggregates during fit and apply during transform if target column is all nan

[lschilders]

As a user
I want to have a fallback for RollingAggregatesAdder
So that I can be robust against missing data during inference

🌍 Background

Important

Changes should be done by forking OpenSTEF 4.0.0 tracking branch (release/v4.0.0) and then make a PR back onto release/v4.0.0 branch.
Please read the contributing guide before you start.

❗Priority (What if we don't do this?/Are there any deadlines? etc.)

Definition of Done:

✅ Acceptance criteria

• Aggregates are computed on the target column during fit and saved in the transform
• in case the target column is empty during predict, transform forward fills these saved values

📄 Documentation criteria:

Interfaces with other products and services

Update relevant confluence pages

🧪 Test criteria:

Unit tests written and passed

⌛ Dependencies:

N/A

🚀 Releasing:

N/A

Other information: