Allow running ML and Apriltags on a single camera

[TheTripleV]

If a color camera is plugged in, I think both object detection and apriltag detection should be allowed to occur at the same time.

I know almost nothing of the PhotonVision internals so a couple approaches I think are possible:

1. Add a new AprilTagPlusObjectDetectionPipeline

Pros:

• Quickest to implement?

Cons:

• The 2 pipelines are hardcoded (Apriltag, Object). There'll be no option to swap in Aruco, Aruco Nano.
• The 2 pipelines will run sequentially, increasing latency (and probably halving framerate) Can 2 persistent threads be created to feed the apriltag detector and the NPU in parallel?

2. Clone the video source using v4l2loopback

Pros:

• No change to most of PhotonVision pipelines. The won't know anything's different.
• 2 pipelines can't both control camera settings so the dummy input will need to be marked as virtual and it's inputs have to be grayed out in the UI.

Cons:

• Allowing a loopback device to set parameters and resolution might be hard to do since the v4l2loopback consumers don't have write access to the producer.

3. Make a copy of the cvSink Source

Pros:

• No change to most of PhotonVision pipelines. The won't know anything's different.
• 2 pipelines can't both control camera settings so the dummy input will need to be marked as virtual and it's inputs have to be grayed out in the UI.

Cons:

• cscore has a copySource but it's not documented so idk if it would work here

[mcm001]

I think copying the cvSink and/or creating a new DuplicateOfAUSBCameraFrameProvider would be the lowest friction way to do this? I dont think ive ever heard of anyone using copySource in cscore before so you should mess with that. Any way you go though youll need a controller/subscriber model where one pipeline can control camera hardware settings and the other isn't allowed to