Summary

This PR fixes two Responses API mapping issues in the gateway conversion layer:

1. function_call linkage now prefers call_id over opaque id when constructing Chat Completions tool_calls[].id.
2. Responses text.format is now mapped to Chat Completions response_format (including json_schema shape conversion).

Why

• In multi-turn tool-call chains, function_call_output links by call_id. Preferring id could create mismatches and orphaned tool outputs.
• Structured output requests that use Responses-native text.format (e.g. responses.parse) were not propagated into chat response_format, causing unconstrained plain-text output in provider-mapped flows.

Changes

• Added resolveFunctionCallLinkId() and switched tool-call ID selection to use call_id first.
• Added mapTextFormatToChatResponseFormat() to convert Responses text.format into Chat response_format, including json_schema mapping to { type: "json_schema", json_schema: { ... } }.
• Added regression tests in openai-chat-to-responses-converters.test.ts:
  • prefers call_id over id for function calls
  • maps Responses text.format json_schema into chat response_format

Validation

• corepack yarn jest --config packages/jest.config.ts packages/__tests__/llm-mapper/openai-chat-to-responses-converters.test.ts
• corepack yarn jest --config packages/jest.config.ts packages/__tests__/llm-mapper