Event-loop stalls occur without CPU spikes

Event-loop can stall without dramatic CPU increases due to GC pauses, synchronous I/O waiting on disk/network, or contention in native code. Container CPU metrics can mask per-process behavior on multi-core machines.

Track process-level CPU (user and system time separately), thread-level CPU where possible, and pair with event-loop utilization (ELU) metric to detect stalls that don't show as CPU spikes.