How can you implement A/B testing for a new product recommendation algorithm in an ecommerce application deployed behind a global HTTP(S) load balancer to evaluate its impact on sales through randomized user allocation?