This avoids pessimizing concurrency on all machines due to e.g. the macOS
machine having high memory usage across the board due to 16K page size.

This also adds max_rss to test-unit and test-c-abi since those tend to eat a
decent chunk of memory too.