Unify variance estimation approaches

[dill]

At the moment, there is dsm.var.prop and dsm.var.gam to estimate variance of abundance estimates. This is confusing, it would be better to include uncertainty from the detection function by default when possible and inform the user which is happening. (There's also a lot of code duplication.)

Also need to implement parametric bootstrap method as suggested in Wood (2006).

[dill]

Also worth making the output from these functions much tidier (tidyr?) -- at the moment it's mostly incomprehensible garbage...

[dill]

moving this to 2.4.x release as it will break a lot of code.

[dill]

Probably best to do this as a UI change to wrap the two existing functions, that will break minimal stuff.

Eventually one could merge them completely, as there is a lot of overlapping code.

For posterior sampling, I'd think a function like dsm_posterior_samples(varprop_result, n_samples=1000, newdata=predgrid) would give you what you want? (It's just a wrapper around rmvn and then some matrix multiplication/application of the inverse link function.)