Extending chain representations

Hi,

I had misunderstood some of the goals of this package, thankfully @cpfiffer got me straightened out. I'm looking into using this as an interface for populating SampleChains.

The docs seem mostly oriented toward people building new samplers, and not so much for people building new ways of representing chains. So I have lots of questions...

1. I'd like to avoid getting all of the samples and then calling bundle_samples. It seems like I should be able to instead overload save!! and then have bundle_samples be a no-op. Is that right?
2. Cameron mentioned a resume function that can pick up sampling where it left off, without needing to go back through the warmup phase. But I don't see it in this repo. Where can I find it?
3. Where are things like per-sample diagnostics stored? Divergences for HMC, that sort of thing.
4. Do you have any examples of using this with log-weighted samples? I need these for importance sampling, VI, etc.

I'm sure I'll have more to come, but this will get me started. Thanks :)

Start by reading the documentation and searching the repository for save!!, bundle_samples, resume, and the existing chain interfaces. Turn the questions about diagnostics, resuming, and log-weighted samples into a concrete scope; the work is done when the supported representation behavior and relevant examples are clearly defined.