[x] accept negative binomial dispersion parameters in two forms: scalar and vector with one parameter per condensed variable
[x] accept observed data streams
[x] compute and return (minus twice) the negative binomial log likelihood
[x] tell TMB what parameters are arguments to the objective function, via the map argument to MakeADFun
[x] introduce an option for not computing the likelihood (currently this happens when length-zero observed data vector is supplied, but I don't know if this is the best solution)
specs: https://canmod.net/misc/flex_specs#mathematical-description-0.2.0