[Feature] Support hardware accelerated video encoding and storage

[tvercaut]

For real-time imaging devices, it is often key to be able to record video data in high-quality while keeping the file size manageable. In the open source community, video encoding is typically done through ffmpeg. Ffmpeg indeed provides many options to tailor the encoding to the specific task at hand. It also supports hardware-accelerated encoding through nvenc, intel quicksync and others. Yet, making proper use of it is not straightforward. Several bindings exist for ffmpeg (e.g. https://github.com/kkroening/ffmpeg-python) but as far as I am aware none support encoding from GPU buffers directly (i.e. without a copy to the CPU).

Some interesting notes about being able to encode GPU buffers through ffmpeg+nvenc without going back to the CPU can be found here: https://stackoverflow.com/questions/49862610/opengl-to-ffmpeg-encode

It would be great if such a use case could be covered in MONAI Stream.

Keeping ffmpeg in the loop would enable the implementation of a graceful fallback to CPU-based encoding but if tapping directly into nvenc is easier, this would already be a strong addition to the library.

There is an interesting code sample here about using nvenc directly with GPUMat from opencv: https://github.com/zteffi/OpenCV-GpuMat-NVIDIA-VIDEO-SDK-Encoder-Sample

Additionally, inline with #74, it would be great to expose such hardware-accelerated video-writer using a simple functional API similar to how it's done in OpenCV's VideoWriter: https://docs.opencv.org/5.x/dd/d43/tutorial_py_video_display.html

[aihsani]

@anishag14 can you triage and scope this? It looks like a candidate past 0.1.0?