Is your feature request related to a problem? Please describe. 📝
For lower-latency, I would prefer if the API was streaming pure bytes instead of creating a DxgiDuplicationFrame object every time. It would be better if it avoids the copy step to NumPy as well.
Describe the solution you'd like 🔑
Stream bytes back from PyO3 instead of the additional indirection via the Python function call, and then an object creation step.