Simulate a real AI coding workflow and measure how stable the node stays during peak traffic.
Runs model listing, 5 non-stream samples, and 1 streaming response check by default.