Design a Streaming Analytics Platform

Design a managed streaming analytics platform in the style of Databricks Structured Streaming. Customers define transformations over event streams, maintain stateful aggregations with checkpoints, and query near-real-time tables.

In scope: micro-batch execution over a batch engine, offsets and checkpoints, state stores and watermarks, exactly-once into transactional tables, restarts and scaling, schema changes, and monitoring. Out of scope: building the message broker and the table format themselves.