Duplicate values in normalized trade names and aliases

eg ERGOCALCIFEROL (Vitamin D)

```json
{
"query": "ERGOCALCIFEROL",
"warnings": [],
"match_type": 80,
"therapy_descriptor": {
"id": "normalize.therapy:ERGOCALCIFEROL",
"type": "TherapyDescriptor",
"label": "ergocalciferol",
"description": null,
"alternate_labels": [
"Calciferol",
"Oleovitamin d, synthetic",
"(5Z,7E,22E)-(3S)-9,10-seco-5,7,10(19),22-ergostatetraen-3-ol",
"Vitamin d2",
"Viosterol",
"D2, Vitamin",
"NSC-62792",
"Vitamin D 2",
"APRD00426",
"(5Z,7E,22E)-(3S)-9,10-secoergosta-5,7,10(19),22-tetraen-3-ol",
"Activated ergosterol",
"Ergocalciferol",
"vitamin D2",
"NUTR00005",
"Vitamina D2",
"Oleovitamin D2",
"Ergocalciferolum",
"Vitamin D2",
"Ercalciol",
"(3β,5Z,7E,22E)-9,10-secoergosta-5,7,10(19),22-tetraen-3-ol",
"Ergocalciférol",
"Viosterol in oil"
],
"extensions": [
{
"type": "Extension",
"name": "trade_names",
"value": [
"Ergocalciferol",
"Calciferol",
"MVI Pediatric",
"Calel-D",
"Calcidol",
"Vitamin D2",
"Vita Numonyl",
"Vitamin D2 (obsolete)",
"Deltalin",
"ERGOCALCIFEROL CAPSULES,",
"DRISDOL",
"VITAMIN D",
"Drisdol"
]
}
],
"therapy_id": "rxcui:4018"
}
}
```

Start by tracing how the ERGOCALCIFEROL example is normalized and how alternate_labels and trade_names are assembled. Compare the resulting JSON with the source data and identify where duplicate values are introduced; done means the normalized response no longer contains those duplicates. No file or test is named in the issue, so locate the relevant normalization entry point and existing tests first.

Written by the indexing model from the issue text.