Context Module

[whuang37]

Added the first context module for agents to better understand xrblocks. Adaptation from #401

• fixed a bug in screenshotSynthesizer where stacked async GPU readbacks could cause a memory leak when .onAfterRender() does multiple readbacks for the same pending screenshot
• added xb.context which now exposes the first context module xb.context.scene
  • exposes three new data pieces:
    • xb.context.scene.tree which is a cut down semantic tree of objects designed for easy agent readability
    • xb.context.scene.visibleObjects which is a list of all objects within the view of the user including occlusion checked via raycasts
    • xb.context.setOfMark an overlayed image of the user's pov with marks that reference objects for agents to reference
    • Unique ids to reference objects in the scene
  • all 3 are dependent on each other so made into one detector with varying detection and enable settings
  • added a new demo for this data demo/context to show this data

Video demo here

[ruofeidu]

Thank you William for filling in the xb.context gap!!

[ruofeidu]

document (0, 0) as bottom-left somewhere?
How was the Gemini object detector returning coordinates so far? like OpenCV or graphics?

[whuang37]

just changed this to origin from top left nromalized. Thats what object detection and faces in the repo currently do