add function for approximate error estimate in PDF fit

[pavoljuhas]

Support error estimates when fitting PDF data that have no ESDs.
Perform fit on a Nyquist grid and estimate uniform standard errors
as sqrt(sum((yobs - ycalc)**2) / (N - P)).

[ycexiao]

@pavoljuhas @sbillinge, I am not sure if I get it right. I am thinking about adding

def approximate_error(self, N, P):

in PDFContribution class. N and P are needed since it is not available at the Contribution level. It is convenient to define the function in the recipe stage, but there is no PDFrecipe class. We can also take the length of the profile as the default for N and 0 as the default for P, since it is not an accurate estimation.