Add response.done message handling for Realtime API error reporting

[jamesrochabrun]

Summary

This PR adds proper handling for response.done messages in the Realtime API, enabling applications to properly surface critical API errors like quota exhaustion and authentication failures.

Problem

Previously, the Realtime API silently swallowed response.done messages, which contain crucial error information. When API errors occurred (e.g., insufficient_quota, invalid_api_key), they were logged as "Unhandled message type" but never surfaced to consuming applications, making debugging nearly impossible.

Solution

1. New Message Type

• Added responseDone(status: String, statusDetails: [String: Any]?) case to OpenAIRealtimeMessage enum
• Captures response completion events with full status details

2. Response.done Handler

• Implemented dedicated handler in OpenAIRealtimeSession for response.done messages
• Parses error details from status_details.error structure
• Logs error code and message for debugging

3. Improved Logging

• Upgraded unhandled message logs from .debug to .warning for better visibility
• Added full JSON logging for unhandled messages to aid debugging

Error Format

The handler correctly parses OpenAI's error structure:

{
  "type": "response.done",
  "response": {
    "id": "resp_xxx",
    "status": "failed",
    "status_details": {
      "error": {
        "code": "insufficient_quota",
        "message": "You exceeded your current quota..."
      }
    }
  }
}

Breaking Changes

None - this is a purely additive change.

Testing

Tested with:

• ✅ Insufficient quota error (proper error surfacing)
• ✅ Successful responses (no side effects)
• ✅ Invalid API key error
• ✅ Normal conversation flows

Impact

Applications using SwiftOpenAI Realtime API can now:

• Detect and display quota/auth errors to users
• Handle critical errors gracefully (e.g., disconnect on auth failure)
• Provide meaningful error messages instead of silent failures

---

🤖 Generated with Claude Code