This PR corrects annotation errors and cleans up extracted-text fields. Changes fall into two groups:

1. **Annotation corrections**: rows where a gold field value did not match the citation string (missing values explicit in the reference, values in the wrong field, or incorrect values).
2. **Text-field cleanups**: normalization of the extracted legal text columns (`full_text`, `absatz_text`, `Gesetzbuch /Norm`, `law_title`).

Row count and schema are unchanged (2945 rows, 22 columns).

## Encoding

Compact citations of the form `Nr. 1a` / `Abs. 1a` are encoded against the statute hierarchy: where a provision is itself numbered `1a`/`4b`, the value is kept whole (`Nummer="1a"`, empty `Buchstabe`) and where a provision has an `Absatz 1` containing a `Buchstabe a`, it is split accordingly.

## Acknowledgement

The corrections combine an internal review with an independent audit reported by Paul Permantier (LexGraph, https://lexgraph.de). Both reviews were cross-checked before merging the corrections.