fix: preserve tool_calls and tool results in Chat→Responses API conversion

[chaizhenhua]

Summary

• Fix Chat Completions → Responses API request conversion to include full message history (assistant tool_calls + tool results), not just the last user message
• Fix Claude API adapter to convert tool_calls and tool_call_id fields when forwarding to Anthropic

Problem

When a client sends a Chat Completions request with tool call history:

{"messages": [
  {"role": "user", "content": "What is the weather?"},
  {"role": "assistant", "tool_calls": [{"id": "fc_xxx", "function": {"name": "get_weather", "arguments": "{}"}}]},
  {"role": "tool", "tool_call_id": "fc_xxx", "content": "{\"temp\": 70}"}
]}

CCProxy's _build_responses_payload_from_chat_request only extracted the last user message as input, dropping the assistant's tool_calls and the tool result entirely. The upstream LLM never saw the tool result, so it re-called the same tool in an infinite loop.

Fix

1. openai_to_openai/requests.py — _build_responses_payload_from_chat_request

Convert ALL non-system messages to Responses API input items:

• user → {"type": "message", "role": "user", ...}
• assistant with text → {"type": "message", "role": "assistant", ...}
• assistant with tool_calls → {"type": "function_call", "call_id": ..., "name": ..., "arguments": ...} per call
• tool → {"type": "function_call_output", "call_id": ..., "output": ...}

2. claude_api/adapter.py — _convert_openai_to_anthropic

Add handling for:

• Assistant messages with tool_calls → Anthropic tool_use blocks
• Tool role messages with tool_call_id → Anthropic tool_result blocks

Test

curl http://localhost:8000/codex/v1/chat/completions \
  -H "Authorization: Bearer sk-ccproxy" \
  -d '{"model":"gpt-5.4","messages":[
    {"role":"user","content":"Weather in Tokyo?"},
    {"role":"assistant","content":null,"tool_calls":[{"id":"fc_xxx","type":"function","function":{"name":"get_weather","arguments":"{\"location\":\"Tokyo\"}"}}]},
    {"role":"tool","tool_call_id":"fc_xxx","content":"{\"temp\":70,\"condition\":\"Sunny\"}"}
  ],"tools":[{"type":"function","function":{"name":"get_weather","description":"Get weather","parameters":{"type":"object","properties":{"location":{"type":"string"}}}}}]}'

Before: LLM re-calls get_weather (tool result was dropped)
After: LLM summarizes "It's 70°F and sunny in Tokyo"