[Workers AI] Correct model reasoning controls and schemas

[KastanDay]

Summary

Correct the reasoning controls in 11 Workers AI model input schemas and explain how Chat Completions and Responses select them. This updates the generated sync-input.json endpoints as well as the model pages: Gemma uses high/none with reasoning off by default, GLM-4.7 exposes its thinking toggle, and GPT-OSS documents mandatory reasoning with a medium default. Compatibility aliases are explained separately from canonical efforts.

Token limits and all non-reasoning model metadata are unchanged. Coordinate the model-schema endpoint release before the next automated catalogue refresh so it retains these corrections.

Validation: full documentation build (9,029 pages), Astro and worker checks, and the generated Gemma sync-input.json pass. Compared all 11 model files against the base to confirm only reasoning fields changed.

Related binding types: workerd #7437.

Documentation checklist

• The change adheres to the documentation style guide.