Cpu usage semantics

[homm]

This PR ports the frequency/usage calculation fixes from #59.

Summary

It changes how usage tuples are calculated:

• sampled CPU/GPU usage is now based on IOReport active residency ratio, not scaled by frequency;
• CPU cluster frequency is reported as the highest active core frequency in the sample;
• CPU cluster usage is normalized to that reported peak active frequency;
• idle CPU clusters now remain 0 MHz / 0% instead of being forced to the minimum DVFS frequency.

The JSON shape is unchanged: ecpu_usage, pcpu_usage, and gpu_usage are still [freq_mhz, usage].

Why

In the previous implementation, calc_freq scaled usage by avg_freq / max_freq, and the result was exposed through fields such as pcpu_usage:

usage = active_residency / total_residency * avg_freq / max_freq

That mixes occupancy with frequency. Under Low Power Mode, thermal limits, or any workload where frequency and occupancy diverge, a fully busy core can look partially idle just because it is not running at the theoretical maximum frequency.

Issue #46 correctly pointed out that the reported value was misleading. The problem was likely not that one CPU cluster was being dropped, but that the metric itself under-reported occupancy by folding frequency into usage.

This PR first computes usage from residency, then aggregates CPU cluster values as a compact "effective load at shown active frequency" summary.

Examples

All examples use a 10-core P-cluster with min_freq = 1250 MHz and max_freq = 4500 MHz.

All idle

  10 cores: 0% @ 0 MHz

Old TUI: P-CPU 0.0% @ 1250 MHz
New TUI: P-CPU 0.0% @ 0 MHz

Why: if the cluster has no active residency in the sample window, there is no observed active frequency to report.

One active core

  1 core: 100% @ 3600 MHz
  9 cores: 0% @ 0 MHz

Old TUI: P-CPU 8.0% @ 1250 MHz (freq = max(3600 / 10, 1250))
New TUI: P-CPU 10.0% @ 3600 MHz

Why: one of ten cores is fully occupied, and the active work is running at 3600 MHz.

One active core, idle cores still report frequency

  1 core: 100% @ 3600 MHz
  4 cores: 0% @ 3600 MHz
  5 cores: 0% @ 0 MHz

Old TUI: P-CPU 8.0% @ 1800 MHz
New TUI: P-CPU 10.0% @ 3600 MHz

Why: idle cores may still carry a reported frequency, but the cluster summary should describe active work.

One peak core plus low-frequency active cores

  1 core: 100% @ 3600 MHz
  4 cores: 80% @ 1000 MHz
  5 cores: 0% @ 0 MHz

Old TUI: P-CPU 15.1% @ 1250 MHz (freq = max((3600 + 1000 * 4) / 10, 1250))
New TUI: P-CPU 18.9% @ 3600 MHz

Why: the summary keeps the observed peak active frequency visible, while lower-frequency work contributes proportionally less to effective load at that frequency.

All cores fully active under frequency limits

  10 cores: 100% @ 2500 MHz

Old TUI: P-CPU 55.6% @ 2500 MHz
New TUI: P-CPU 100.0% @ 2500 MHz

Why: the cluster is fully occupied. A lower frequency caused by thermal or power limits should not make fully busy cores look idle.