Epic 1: Structured LLM Output (M27)

[bug-ops]

Problem

All LLM responses are raw String. Parsing structured data is ad-hoc (serde_json::from_str on fenced blocks). No way to force JSON schema compliance.

Solution

Add chat_typed<T>() to LlmProvider trait + provider-level response_format support.

Issues

• feat(llm): add chat_typed<T> method to LlmProvider trait #456 feat(llm): add chat_typed<T> method to LlmProvider trait size/M
• feat(llm): OpenAI structured output via response_format #457 feat(llm): OpenAI structured output via response_format size/S
• feat(llm): Claude structured output via forced tool use #458 feat(llm): Claude structured output via forced tool use size/S
• feat(llm): Extractor utility for typed data extraction #459 feat(llm): Extractor utility for typed data extraction size/S

Verification

Write a test that calls provider.chat_typed::<MyStruct>(messages) and gets a typed result.