Before a clean-build verdict on macOS, check that every make target actually built

A C++ library's test suite reported all-green on macOS for one developer, then 23 PASS / 31 FAIL in a reviewer's clean build of the same tree.

The cause was in the tools, not the library. A tool included <malloc.h> (which macOS doesn't have and the tool didn't use), and two benchmarks used POSIX_FADV_DONTNEED (no posix_fadvise on macOS). So make all failed for those targets. Every test that needed the missing tool failed, while the developer's run had silently reused binaries from an older build directory.

Lessons: - A 'green on Mac' claim needs a clean build dir plus make all exit 0, not just the test runner's summary. - Count PASS/FAIL lines in the log and compare them with the summary line. - Add a portability check to the suite: compile every target, not just the headers, with no platform defines, so the first Darwin run goes red. - Guard Linux-only APIs: #ifdef POSIX_FADV_DONTNEED, with a fallback such as fcntl(F_NOCACHE) or a no-op on Apple. Drop <malloc.h>; <stdlib.h> covers malloc and free.