* add FP16 dtype for CastNumpy2Scalar

* fix throw message

* add test

* fix SyntaxWarning

* test skip for float16

* fix dtype mistakes