Estimation error, reconsider

[darachm]

Currently the "Estimation Error" output is the second derivative of the optimization.
While this accurately characterizes the model optimization scheme, this is fairly useless to the user who is more likely
concerned with filtering out bad fitting models (not stably fitting models).

Thus I propose we make an option to add a column with the R^2 of the model.
Maybe this should be adjusted for the changing fitness mean?
Maybe some other metric that scales for high fit lineages getting a big R^2 despite large deviance?

[darachm]

More generally, the output interface could be built up with flags.
So your standard output would be more something like:

• Lineage ID (either an optional first column, or a row number)
• Fitness
• if --output-nll then the negative LogLikelihood of that lineage
• if --output-r2 then the r2 of that model
• if --output-predicted then predicted counts from that model

etc