Implement `posterior::as_draws()` for cmdstanr objects

[mjskay]

Per the conversation at stan-dev/posterior#183, now that {posterior} is on CRAN it would be wonderful if cmdstanr's objects could have an implementation of posterior::as_draws() so that users would not have to remember what function to call to get draws objects out of models when moving between packages. Perhaps this could be a simple wrapper around the $draws() method? Thanks!

[jgabry]

Sorry for the delay on this. I'm going to tackle this soon.

Is the idea to add each of the as_draws_* functions (e.g. as_draws_array(), as_draws_df(), etc.) or just the single as_draws()? Judging from the conversation in stan-dev/posterior#183 it seems like I should implement all the as_draws_* functions, right? However, just having the as_draws() function could be sufficient since it gives the user a draws object that they can then convert to other formats using the standard functions in posterior.

I'm open to either option. @rok-cesnovar @mjskay @paul-buerkner any preference?

Also it sounds like from the discussion in stan-dev/posterior#183 we should have arguments inc_warmup and variables, so I can add those (which can just be passed to CmdStanR's draws() method).

[mjskay]

I guess it depends on if you either (1) have need for extra parameters like brms does or (2) have more efficient implementations of those functions than the generic implementations in posterior will be. If no to both, I think you can just add an as_draws() implementation that converts to the closest format, and the generics for posterior::as_draws_XXX() will call down to as_draws() anyway, so users will still be able to use as_draws_XXX() directly on cmdstan objects.

[jgabry]

Thanks! That all makes sense.

[jgabry]

Finally found time to make a PR for this: #581

[jgabry]

Since this is just a wrapper around fit$draws(), I ended up just using ... as the only argument, which allows any of the arguments supported by fit$draws().