Function populate_run_pooled_iterations_per_cpu_sec 

pub fn populate_run_pooled_iterations_per_cpu_sec(stats: &mut ScenarioStats)

Inject the run-level POOLED iterations_per_cpu_sec Rate’s two Counter components into stats.ext_metrics, summed across the cgroups that have measured on-CPU time — the cross-cgroup re-pool axis. Rather than routing the per-cgroup efficiency through AssertResult::merge’s worst-by-polarity ext_metrics fold (which picks the WORST cgroup’s value, not Σ, and has no derive post-pass), this reads the already-merged stats.cgroups vec directly: iterations_per_cpu_sec = Σtotal_iterations / Σ(total_cpu_time_ns/1e9) over cgroups with total_cpu_time_ns > 0 — the per-cgroup CgroupStats::iterations_per_cpu_sec re-pooled, NOT a mean of per-cgroup ratios, NOT the worst single cgroup.

MUST run at the eval layer AFTER the cgroup-bearing merges (every merge that contributes a CgroupStats, so stats.cgroups holds every per-cgroup entry) and BEFORE the sidecar write. The trailing monitor-verdict merge at the eval layer merges an inconclusive() carrying empty stats (no cgroups, no ext keys), so it is safe to run after this. If component injection ever moved BEFORE a cgroup-bearing merge, that worst-by-polarity fold would min/max these Counter keys into single-cgroup scalars, silently corrupting the pooled sum.

A cgroup with total_cpu_time_ns == 0 (schedstat unavailable, or num_workers == 0) is EXCLUDED from BOTH sums — mirroring the per-cgroup CgroupStats::iterations_per_cpu_sec None-on-zero (total_cpu_time_ns > 0 implies num_workers > 0, so the one predicate covers both). Crediting an unmeasured cgroup’s iterations against the measured cgroups’ CPU-seconds would overstate cohort efficiency — the silent-wrong-answer this gate prevents. Both components are inserted both-or-neither (the derive_rate_metrics co-location invariant), only when the summed MEASURED on-CPU time is > 0 (every cgroup unmeasured ⇒ no rate). The ns→s /1e9 is applied ONCE here on the summed ns (not per-cgroup, to avoid repeated float rounding), since derive_rate_metrics is a bare num/den. total_iterations_pooled is a DISTINCT ext-only key, not the typed total_iterations (skipped from ext_metrics; it folds cross-RUN as a MEAN — a display average — while a Rate numerator must SUM-fold so Σnum/Σdenom re-pools, so one shared key cannot carry both folds). Because it sums only MEASURED cgroups, it is ≤ the merge-summed typed total_iterations (which includes any zero-cpu-time cgroups), and equals it unless an excluded zero-cpu-time cgroup carried iterations>0.