Better data error type deduction in data modeling

Currently, the default error type used in data modeling fit is Poisson which is problematic when we have weighted events. A better default would be to deduce the data error type automatically from the dataset.

See ROOT documentation for ErrorType in plotOn: https://root.cern.ch/doc/master/classRooAbsData.html#a00a476834b0f2d246f85da34a915fcd9