Function amd_order_substages 

pub fn amd_order_substages(
    pattern: &CscPattern<'_>,
    opts: &AmdOptions,
) -> Result<(Vec<i32>, AmdSubstages), OrderingError>

Compute an AMD ordering and report a per-sub-stage timing breakdown.

Behaves identically to amd_order_full but additionally returns an AmdSubstages split of the wall-clock time across workspace::new, run_elimination, and finalize_permutation. Diagnostic-only — production callers should keep using amd_order / amd_order_full. The Instant::now() calls add at most ~100 ns vs the un-profiled path, but the API surface is intentionally separate to avoid polluting the stable contract-conforming function.