Fix docs-to-implementation discrepancies

[mintlify]

Summary

Compared documentation against the actual source code in packages/ and fixed several discrepancies and gaps:

Discrepancies fixed

• Mistral reasoning support: Updated capabilities table from "No" to "Partial" — the adapter parses and streams reasoning content (thinking blocks) but does not support the reasoning.effort config parameter
• Missing gemini-3-pro model: Added to Google GenAI supported models list and comparison table
• Mistral provider options: Added missing parallelToolCalls and promptMode to both the Mistral provider page and the configuration summary page

Missing documentation added

• Utility functions in types.mdx: Added resultToMessage(), assistantMessage(), and getProviderMetadata() documentation
• Reasoning metadata types: Added OpenAIReasoningMetadata, AnthropicReasoningMetadata, and GoogleReasoningMetadata type documentation with usage examples to each provider page
• Reasoning constraints: Documented OpenAI sampling parameter restrictions when reasoning is enabled, and Anthropic's validation rules (temperature, topP, toolChoice, topK constraints)

Improvements

• Expanded Mistral generate options from a flat list to a structured table with types and descriptions
• Added Google GenAI model comparison table entries for Gemini 3.1 Flash Lite