Add test cases for u16, i16, and f32 pixel types

[chung-leong]

32-bit integer types cause problems as their max values cannot be accurately stored in a 32-bit float.