What is TCP Slow Start and how does this mechanism affect the latency of the first requests to the backend? Why is it critical to use persistent connections (Keep-Alive / Connection Pooling) for inter-service communication within a data center?