Variant winner with guardrail

A growth experiment ships only if the variant wins the primary metric without wrecking a guardrail. Each arm is given as [conversions, visitors, unsubscribes] with visitors >= 1. Declare the variant the winner (return 1) exactly when both hold, using integer arithmetic only: (1) its conversion rate is strictly higher: variant_conv * control_visitors > control_conv * variant_visitors; and (2) its unsubscribe rate does not exceed control's by more than margin_bps basis points: 10000 * (variant_unsub * control_visitors - control_unsub * variant_visitors) <= margin_bps * control_visitors * variant_visitors. Otherwise return 0. Example: control = [50, 1000, 10], variant = [70, 1000, 12], margin_bps = 30 returns 1.