Bug Description

When using third-party OpenAI-compatible APIs (like Alibaba Cloud Bailian/Qwen) with the `openai-completions` API type, tool calling does not work. The model outputs tool commands in text format (e.g., ```bash code blocks) instead of actually invoking the tools.

Steps to Reproduce

1. Configure a third-party OpenAI-compatible API provider (e.g., Alibaba Cloud Bailian):
   ```json
   {
   "models": {
   "providers": {
   "bailian": {
   "baseUrl": "https://coding.dashscope.aliyuncs.com/v1",
   "apiKey": "your-api-key",
   "models": [
   {
   "id": "qwen3-coder-plus",
   "api": "openai-completions",
   "compat": {
   "supportsTools": true
   }
   }
   ]
   }
   }
   }
   }
   ```

2. Run: `openclaw agent --agent main --local -m "Use exec tool to run echo test"`

3. Observe that the model outputs ```bash exec --command="echo test"``` in text instead of actually calling the tool.

Verification

Direct curl request to Bailian API works correctly with tool calling:
```bash
curl -s "https://coding.dashscope.aliyuncs.com/v1/chat/completions"
-H "Authorization: Bearer $API_KEY"
-H "Content-Type: application/json"
-d '{
"model": "qwen3-coder-plus",
"messages": [{"role": "user", "content": "Check weather in Beijing"}],
"tools": [{"type": "function", "function": {"name": "get_weather", "parameters": {...}}}]
}'
```
Returns: `"finish_reason":"tool_calls"` with proper tool call response.

Environment

• OpenClaw: 2026.3.8 (3caab92)
• Node.js: v24.14.0
• Platform: Linux (WSL2)

Root Cause

The `openai-completions` API type does not properly send tool definitions to the model, even when `compat.supportsTools: true` is configured.

Suggested Fix

1. Add a new API type `openai-chat` that explicitly uses `/v1/chat/completions` endpoint with proper tool calling support.
2. Or fix `openai-completions` to properly handle tool definitions for OpenAI-compatible APIs.