Mistral catalog omits callable IDs returned by GET /v1/models

Problem

The Mistral provider in api.json currently has 34 entries. Mistral's authenticated GET /v1/models returned 56 active entries on 2026-08-27, with 36 callable IDs absent from models.dev.

Mistral's API reference documents the model list, including id, capabilities, max_context_length, aliases, and deprecation fields:
https://docs.mistral.ai/api/endpoint/models

Missing IDs

codestral-2508
codestral-embed
codestral-embed-2505
glm-5-2
labs-leanstral-1-5
labs-leanstral-1-5-1
magistral-small-latest
ministral-14b-2512
ministral-14b-latest
ministral-3b-2512
ministral-8b-2512
mistral-code-agent-latest
mistral-code-fim-latest
mistral-code-latest
mistral-embed-2312
mistral-medium
mistral-medium-3
mistral-medium-3-5
mistral-medium-3.5
mistral-moderation-2603
mistral-ocr-2512
mistral-ocr-3
mistral-ocr-3-0
mistral-ocr-4
mistral-ocr-4-0
mistral-ocr-4-1
mistral-ocr-latest
mistral-vibe-cli-fast
mistral-vibe-cli-latest
mistral-vibe-cli-with-tools
voxtral-mini-2602
voxtral-mini-realtime-2602
voxtral-mini-realtime-latest
voxtral-mini-transcribe-realtime-2602
voxtral-mini-tts-2603
voxtral-small-2507

Some are aliases or dated IDs for underlying models already represented in models.dev. They still matter to consumers because they are accepted model identifiers and may be echoed back in API responses. base_model should allow those entries to inherit shared facts while retaining the exact provider ID.

The embedding, moderation, OCR, transcription, and speech entries may expose the same non-chat schema limitations discussed in #572, but current chat/coding aliases are missing independently of that.

Start in api.json by comparing the Mistral provider entries with the documented GET /v1/models response and the 36 IDs listed here. Add the missing callable identifiers, using base_model where shared facts can be inherited while preserving each exact provider ID; done means the catalog represents those accepted IDs without omitting the non-chat entries.