fix: read error field from top-level data, not message object

[dicksontsai]

Summary

Fixes #505

The AssistantMessage.error field was never being populated because the message parser was reading from the wrong location in the JSON structure.

Problem

The CLI outputs error information at the top level of the JSON, not inside the message object:

{
  "type": "assistant",
  "message": { "content": [...], "model": "..." },
  "session_id": "...",
  "uuid": "...",
  "error": "unknown"  // <-- error is HERE at top level
}

The previous fix (PR #405) incorrectly read from data["message"].get("error") instead of data.get("error").

Solution

Changed the parser to read error from the correct location:

# Before (wrong)
error=data["message"].get("error"),

# After (correct)
error=data.get("error"),

Testing

• Added 3 new test cases for error field parsing:
  • test_parse_assistant_message_with_error - general errors
  • test_parse_assistant_message_with_rate_limit_error - rate limit detection
  • test_parse_assistant_message_with_auth_error - authentication errors
• All existing tests pass
• Passed ruff linting
• Passed mypy type checking

Usage Example

After this fix, applications can properly detect and handle errors:

async for message in client.receive_response():
    if isinstance(message, AssistantMessage):
        if message.error == "rate_limit":
            print("Rate limit hit! Implementing backoff...")
        elif message.error == "authentication_failed":
            print("Invalid API key")
        elif message.error:
            print(f"API error: {message.error}")