Fix issue with variable initial loss for different zref

[alizma]

For the pure Mann fit, the initial loss changes when zref changes, which shouldn't happen. To summarize the discussion of this issue thus far: this is a scaling issue and happens either because the data from DataGanerator are improperly scaled, or because the output of theOnePointSpectra class immediately after initialization are incorrectly scaled. Refer to the discussion in #60 for more details.

[alizma]

This issue is being contained to this branch with the debug trace from our earlier attempt today.