Implementation of theory error

[tgiani]

We need an implementation of theory error to allow for correlations between different datasets. Currently not implemented at all (the flags controlling the theory error in the code do not do anything). Having a different theory covariance matrix for each dataset doesn't make much sense.

[giacomomagni]

@jacoterh so regarding the theory covmat there are 2 issues that can be addressed:

• Multiple entries for the theory covmat i.e. PDFu, MCu, SV ...
• Correlations between different datasets, because for instance PDFu are fully correlated in all datasets.

This is still a bit vague or partially implemented, so with @tgiani we will try to address it in the next weeks.