feat: Add OpenAI-compatible backend for local model support

[weijt606]

feat: Add OpenAI-compatible backend for local model support

Summary / 概述

Add a new openai proposer backend that enables PolyHarness to work with any OpenAI-compatible API endpoint — including locally deployed models via Ollama, vLLM, LM Studio, and similar tools.

新增 openai 后端，支持通过 OpenAI 兼容接口连接本地部署模型（Ollama、vLLM、LM Studio 等），让用户无需云端 API 即可使用 PolyHarness 进行 harness 优化。

Changes / 变更内容

• New file: openai_proposer.py — OpenAI-compatible proposer with full tool loop (file_read, file_write, list_dir, bash)
• Config: Added base_url and api_key fields to ProposerConfig for custom endpoint configuration
• CLI: Added openai to --agent and --backend choices in ph init / ph run
• Dependency: Added openai>=1.0.0 to pyproject.toml
• Docs: Updated README.md and README_CN.md backend tables
• Version: Bumped to 0.1.2

Usage / 使用方式

ph init --agent openai
ph config set proposer.model llama3.3
ph config set proposer.base_url http://localhost:11434/v1
ph run

Testing / 测试

• 121 tests passed
• ruff lint clean