OVRTX semantic AOV capture

Use ovrtx-render-products-and-aovs for product/readback mechanics and extend-ovrtx-semantic-aovs for implementation.

1. Define durable semantic class/instance ownership in USD separately from the visualization. Preserve stable prim identity through composition.
2. Add an explicit product/RenderVar path and typed native builder/decoder. Probe the installed native module's capabilities; fail as unsupported when the semantic builder or decoder is absent.
3. Preserve dtype, dimensions, background/unlabeled value, class/instance mapping, sample time, product identity, and terminal status.
4. Test duplicate labels, missing labels, hidden/excluded prims, remapped prim paths, iterator pages, wrong dtype/shape, terminal errors, and cleanup.
5. Validate a tiny fixture with two known labeled objects and one unlabeled background before a real scene. Require exact pixel IDs at known regions and stable mapping across repeated reads.