Add DINOv3/DINOv2-B world models, phi-space f, pixq encoder, and 29 V ensembles (2026-09-12 campaign)

threading_d0 dynamics + value models, state and pixel