Add TPC-C test under the chaos suite

[vnvo]

What

• 9 tables, 5 transaction types with a fixed distribution: 45% New-Order, 43% Payment, 4% Delivery, 4% Order-Status, 4% Stock-Level
• New-Order: INSERT into order + 5-15 order_line rows in one transaction - tests CDC transaction atomicity
• Payment: UPDATE customer + district + warehouse - pure multi-table UPDATE stream
• Delivery: batch UPDATE across 10 order_line rows + UPDATE customer - tests bulk UPDATE throughput
• Stock updates are high-contention (same rows updated by many writers concurrently) - surfaces lock/retry handling
• Result: we get INSERTs, UPDATEs, and DELETEs across fk-linked tables in realistic proportions, all within "real" transactions. That's exactly the hardest thing for a CDC engine to get right.