mesa: Fix _mesa_float_to_unorm() on 32-bit systems.

This fixes long vs long long issues in _mesa_float_to_unorm on 32-bit systems. See the last commit message for more details.

+@mattst88