validator: Arithmetic on 16, 8, 64bit types only permitted by associated capabilities

16bit integer arithemtic is only permitted if Int16 capabilities declared (explicitly or implcitly). To begin with, this was guarded by the ability to just *declare* a 16bit integer type. But extension SPV_KHR_16bit_storage adds capabilities that allow you to declare 16bit type and only do width-conversions.

Similar for 16bit float.