feat(models): add fast-tier model variants

[RhysSullivan]

Summary

• extend the model schema so provider-specific request options can be encoded directly in model TOMLs
• add fast-tier variants for gpt-5.4 and claude-opus-4-6, including OpenCode Zen overrides so harnesses can expose them as distinct priced models
• encode the documented fast-tier transport options (serviceTier = \"priority\" and speed = \"fast\") in the catalog instead of relying on UI-local toggles

Validation

• bun validate

Notes

• gpt-5.4-fast uses OpenAI priority processing pricing from the API pricing docs
• claude-opus-4-6-fast uses Anthropic fast mode pricing from the fast mode docs and is marked beta because the feature is still in research preview