Deduplicate and Impute Data

Deutsche Telekom data pipelines receive records with inconsistent fields and repeated entries. Implement a function that normalizes missing values, identifies duplicates using one or more key fields, and separates records whose key is incomplete.

Formal Specification

Given records, a list of dictionaries, and key_fields, a list of field names:

1. Treat an absent field, None, or an empty string as a missing value.
2. Add every field found in any record to every normalized record, using None for missing values.
3. Use the values of key_fields as a composite key.
4. If any key field is missing, place the normalized record in missing_key_records.
5. For valid keys, retain the first occurrence in unique_records. Place later occurrences in duplicates.
6. Return a dictionary with the keys unique_records, duplicates, and missing_key_records.

All non-missing field values are hashable. Preserve each record's values and field names apart from adding missing fields with None.