Allow per-call object detection snapshots and backends

[salmanmkc]

Description

the 3D detector has to patch the shared camera getter and switch the global backend to reuse a captured frame. overlapping calls can also pick up an in-flight result for the wrong request. this is the runDetection({snapshot, backend}) follow-up from #417.

adds per-call inputs and queues explicit requests separately, keeping their camera/depth state while they wait. calls without inputs still share the active detection, and simulator ground truth keeps taking precedence. switches the addon and demo callers over; the larger demo migration stays in #487.

tested in the desktop simulator with real MediaPipe, and manually in the browser with live Gemini. also ran 41 browser cases for queueing, errors and cleanup (Gemini responses mocked in those automated cases). full SlimSAM validation and headset testing remain outstanding.

Type of Change

• Bug fix
• New feature / enhancement
• New demo or sample
• Documentation update

Media / Screen Recordings & Screenshots (If Applicable)

• Simulator Recording: No recording attached.
• Device Recording: Not recorded; physical XR hardware was not tested.

Checklist

• Tested in simulator & device: Verified functionality in desktop simulator and/or physical hardware (where applicable). Desktop simulator only.
• Large Assets ($\ge$ 1MB): Submitted separately to xrblocks/proprietary-assets via jsdelivr CDN. N/A; no large assets added.
• SDK Dynamic Dependencies: All new SDK dependencies are dynamically loaded at runtime. N/A; no new SDK dependencies.
• Security: Confirmed no hardcoded API keys or secrets are committed.