Merged via rebase, thanks @cmeerw. The #elif restructure is exactly the right shape — pulling BSDs out before the implicit Linux fallback fixes the root cause (the previous #elif !defined(_WIN32) /* Linux */ was branch-labeled "Linux" but actually included every non-Windows non-Apple system, which is why sysinfo() was being pulled in on NetBSD).

The OpenBSD-vs-FreeBSD/NetBSD split on HW_PHYSMEM vs HW_PHYSMEM64 is the right call — OpenBSD never grew the 64-bit variant, and on modern 64-bit OpenBSD HW_PHYSMEM returns the full value into our uint64_t anyway. Issue #290 auto-closed with the merge.

Treating all BSD cores as perf_cores matches the Linux path (BSDs don't expose a perf/efficiency distinction) — also correct.