Summary

Simplified error handling for both streaming and non-streaming chat completions to provide clearer, more transparent error messages to users.

Changes

• Streaming errors: Send error messages via SSE chunks with finish_reason: "error" instead of trying to change HTTP status after headers are sent
• Non-streaming errors: Return HTTP 500 with raw error message in JSON response
• Removed error sanitization: Use error.localizedDescription directly for transparency
• Added comprehensive unit tests: 14 test cases covering payload parsing, error formats, and response serialization

Benefits

• ✅ OpenAI-compatible error responses
• ✅ Clearer error messages for debugging
• ✅ Consistent error handling between streaming/non-streaming
• ✅ Robust test coverage for both error and success cases

Testing

All tests pass: swift test --filter CompletionsHandlerTests (14/14 ✅)