Add final MNAR imputation project.

[elisawege]

This pull request adds the final artifacts for the MNAR imputation project.

The project includes:

• final Python prototype
• final DML/Systemds prototype
• missingness generators
• tests
• documentation
• experiment scripts and selected result artifacts

Core Idea:
A rebalanced MICE-like imputation approach for MNAR data, using target-region weighting and similarity-based weighting.

How to run:
cd mnar_project

Diabetes Dataset
python3 src/run_all_experiments.py --input-path data/processed/diabetes_sample.csv

AQ dataset:
python3 src/run_all_experiments.py --input-path data/processed/AQ_clean.csv

Main result:
Compared with standard MICE, the rebalanced approach substantially reduces MNAR-related bias. The final prototype was translated into DML/SystemDS.

[mboehm7]

LGTM - thanks for the patch @elisawege. During the merge, I added the missing license headers, and cherrypicked the dml-scripts and documentation. However, the exploratory experiments are still reachable from this PR.