Cohere catalog omits active models returned by GET /v1/models

Problem

The Cohere provider in api.json currently has 14 models. Cohere's authenticated GET /v1/models?page_size=1000 returned 32 non-deprecated models on 2026-08-27, with 20 callable IDs absent from models.dev.

Cohere documents this endpoint and its endpoints, features, and context_length fields here:
https://docs.cohere.com/reference/list-models

Missing IDs

cohere-transcribe-03-2026
embed-english-light-v3.0
embed-english-light-v3.0-image
embed-english-v3.0
embed-english-v3.0-image
embed-multilingual-light-v3.0
embed-multilingual-light-v3.0-image
embed-multilingual-v3.0
embed-multilingual-v3.0-image
embed-v4.0
parse-v5.0
rerank-english-v3.0
rerank-multilingual-v3.0
rerank-v3.5
rerank-v4.0-fast
rerank-v4.0-pro
tiny-aya-earth
tiny-aya-fire
tiny-aya-global
tiny-aya-water

The list response identifies embedding, rerank, transcription, and chat models by endpoint, so those classifications do not require guessing from the IDs. Cohere also documents the current rerank IDs directly:
https://docs.cohere.com/docs/rerank

The embedding/rerank subset is partly blocked by #572 because the current modalities enum cannot represent vector or rerank output. The chat models and the underlying catalog completeness are independent of that schema gap.

Start in api.json and compare the Cohere provider catalog with the authenticated GET /v1/models response and the linked Cohere documentation. Add the active callable model entries and their documented endpoint classifications, while keeping the embedding and rerank schema limitation in #572 separate. Done means the catalog includes the missing chat models and the provider’s active model list is complete.