t2m: retrain with corrected 6D rotation packing (#411, PR #792 review)

t2m v4 vocab: 13 actions, frame=world, fps=30 (#411)

t2m v4: 30fps world-frame CVAE, absolute decoder, derived-local supervision (#411)

motion library v4: 47 active-window clips / 15 actions with per-action variety (#411)

Add t2m vocab (#411)

Add experimental t2m model (#411)

Add text-to-motion template clip library (#411 MVP) — 10 CMU-derived actions