kernel32: Remove _hread16() as it is not used. Move _hwrite16() next to the other related functions.

It's probably slightly faster too.

wined3d: We only care is a sampler is sampled at all in device_map_vsamplers(), not the specific type.

msvcrt/tests: Load __mb_cur_max and _mbctype dynamically since they may not be available to link against.