From Conversation to Query Execution: Benchmarking User and Tool Interactions for EHR Database Agents