feat(runtime): add openai-responses provider and Responses API routing

[mxyhi]

Summary

• add a new provider alias: openai-responses
• route OpenAI driver to /v1/responses when model format requires Responses API
• allow explicit force format via openai-responses/<model> prefix
• support OPENAI_REASONING_EFFORT (low|medium|high|xhigh) and model suffix fallback
• update docs for provider/configuration usage

Implementation

• drivers/mod.rs: register openai-responses and include it in known provider hints
• drivers/openai.rs: add Responses request/response structs and routing logic
• agent_loop.rs: prepend openai-responses/ to api_model when provider is openai-responses

Tests

• cargo check -p openfang-runtime
• cargo test -p openfang-runtime drivers::openai::tests:: -- --nocapture
• cargo test -p openfang-runtime test_known_providers_list -- --nocapture

[jaberjaber23]

Good feature — OpenAI Responses API for o-series reasoning models. Has merge conflicts. Please rebase onto main.

[mxyhi]

Superseded by #628.

I rebased feat/openai-responses-reasoning-effort onto current main and pushed the updated branch, but GitHub would not reopen this PR. The replacement PR is #628: #628

[jaberjaber23]

@mxyhi saw the rebased #628. The core issue is the request schema doesn't match OpenAI's actual Responses API format. The ResponsesRequest struct uses fields (instructions, input) that the real endpoint doesn't accept, and build_responses_input() converts messages to plain text instead of the structured JSON the API expects. Please verify the schema against https://platform.openai.com/docs/api-reference/responses and resubmit. The driver architecture (routing based on model name, separate complete_via_responses path) is solid, just needs the real API format.