fix(metrics): cpu_stat deltaWaitRunSum uint64 underflow on cgroup counter reset

[wc4440222]

Fix

Fixes #414

Problem

In core/metrics/cpu_stat.go, deltaWaitRunSum is computed with a raw uint64 subtraction:

deltaWaitRunSum := deltaHierarchyWaitSum + stat.cpuTotal - cpu.cpuTotal

When a container's cgroup is recreated (e.g., container restart), stat.cpuTotal can be smaller than cpu.cpuTotal, causing the subtraction to underflow to a near-MaxUint64 value. This produces garbage values for all four wait-rate metrics derived from deltaWaitRunSum.

Fix

Reuse the already-safe deltaCpuUsage variable (computed with a counter-reset guard a few lines above) instead of repeating the raw subtraction:

deltaWaitRunSum := deltaHierarchyWaitSum + deltaCpuUsage

This is consistent with how waitSumDenom is already computed on line 144:

waitSumDenom := deltaWaitSum + deltaCpuUsage

Verification

• go build ./core/metrics/ — pass
• golangci-lint run ./core/metrics/... --timeout=5m --config .golangci.yaml — pass (no warnings)
• Code review by sub-agent — pass