Created by: FrostML

According to 单测规范, it's necessary to update the precision of pad, pad2d, pad_constant_like's unit tests from fp32 to fp64. What's more, fix the data type error caused by pad_value, which is float, but got by double.