win-speed bonus: record + reward how fast a model wins

#2 hard: per-waypoint search BAND in the relative objective

#2: enforce ordered waypoints (stateful waypoint_sequence) + two-path medium + scouted hard

action-sequenced-execution: timeout/attrition overrun is a LOSS, not a draw

hard: enemy_building_spotted interrupt closes the coordinate-blind feedback loop

hard: anchor relative objective on the named enemy building type per direction