The short and byte mixers and prependers just delegate to the int variants. At the LambdaForm level the values has already been implicitly cast to int, so removing these and directly adapting to call the int-based variants reduces a few steps.

This is mainly a cleanup. Most of the generated machinery deal with basic types where byte and short has been widened to int, so this has little actual effect on bootstrap times or other overheads. (Same number of generated classes; a slight reduction in generated MethodTypes, etc.).