use sized deallocation internally for ralloc

The size of the source allocation is known at this point, so reading the
chunk header can be avoided for the small size class fast path. This is
not very useful right now, but it provides a significant performance
boost with an alternate ralloc entry point taking the old size.