Function run_parallel 

pub fn run_parallel<R, F>(work: F) -> R
where
    R: Send,
    F: FnOnce() -> R + Send,

Run work under the active thermal policy, throttling parallelism to the thermal cap only when appropriate.

Wrap a parallel kernel’s body in this. The rules:

• cap ≥ cores (max-perf / --threads ≥ N): run on the full global pool.
• already inside a rayon worker (nested call): run inline on the current pool — never nest-install (avoids surprising thread fan-out / blocking).
• adaptive (default): full width during a burst, throttle once load is sustained ([decide_sustained]).
• adaptive == false: always throttle to the cap (fixed, reproducible).

Throttling means install-ing into the [capped_pool], inside which rayon::current_num_threads() reports the cap — so existing chunkers that size their work to the live thread count auto-scale. Determinism is preserved: the choice of pool changes only how many bands/rows run concurrently, never the per-element math (reductions keep their fixed within-row order), so output is bit-identical regardless of this decision.