resolve num_error, result error. all to float32

[JernKunpittaya]

Clarify difference between num_error and result_error.
-We would call our self.error as result_error as acceptable range where our final result differs from theory, where num_error is the acceptable range that we regard two number as the same.
-Turn out median doesnt need num_error, despite low scale, so we can treat like other functions
-For mode, we hardcode num_error to be 0.01, which may make sense in dataset that has all data different from each other making finding mode, not making sense. To be discussed how to best deal with this num_error for mode case @mhchia

Switch everything to float32. Since torch.tensor() defaults to float32, while numpy defaults to float64, this may cause inequality in comparing number.

[mhchia]

I think it should be good to hardcode num_error for now. If we need to configure it we can easily modify the State to accept another num_error later