Use dtype of RHS in NdArray.data setter

[TakuyaNarihira]

Previously, there was an issue that an array with any dtype is automatically converted to fp32 at the first call of .data property in an instance of NdArray.

a = np.arange(10)  # int64
b = NdArray(a.shape)
b.data = a  # b.dtype becomes float32 

By this change, the setter uses a dtype of RHS as following.

b.data = a # b.dtype becomes int64

There is an exception where zero() or fill(rhs) is invoked if a scalar with a float or an integer <= 2^53 (as value to be filled is maintained as float64) is given.

Additionally, resolved the following;

• Remove unnecessary prints in testing
• Properly raises an exception in function call, and fix min-max quantize affine