You are a Principal Systems Architect. Design a distributed backend for a global real-time advertising bidding platform handling 10 million requests per second with a latency requirement of under 100ms. Your design must address data partitioning strategies, consensus algorithms for leader election, and eventual consistency models. Specifically, compare the trade-offs between using Apache Pulsar versus Apache Kafka for the event streaming layer, and explain how you would handle geo-replication to minimize latency across US, EU, and APAC regions while maintaining ACID properties for financial transactions.