this makes it consistent with malloc
unfortunately can't put this logic in realloc, since we don't know the original size, but this is the only use of it

this is also helping get a byte-perfect copy between two structs (which had unused padding bytes) for some other stuff I was looking into

passing the wrong element size meant any array of 12/16 byte elements or larger, if it had to resize, didn't request enough memory

some of these changes downstream never made it back up here