refactor(ai-openai): redesign model metadata catalogs

[nick-potts]

🎯 Changes

This adds the latest OpenAI models and snapshots across text, audio, image, video, and realtime.

It also replaces the old ai-openai metadata with registry-backed catalogs so there’s a single source of truth for supported models and capabilities, which avoids the silent drift that had already started to show up across types, validation, and exports.

✅ Checklist

• I have followed the steps in the Contributing guide.
  this doesn't exist
• I have tested this code locally with pnpm run test:pr.

🚀 Release Impact

• This change affects published code, and I have generated a changeset.
• This change is docs/CI/dev-only (no release).

Summary by CodeRabbit

• New Features

  • Unified model registry across modalities with additional image, audio, and video model entries.

• Improvements

  • Metadata-driven validation for model options and clearer documented supported models (image, TTS, realtime).
  • Realtime default switched to a newer realtime model.

• Bug Fixes

  • Safer transcription handling with robust fallbacks when upstream responses lack expected fields.
  • Validation now warns or rejects more predictably for unknown/unsupported models.