ppc_loo_* consistency

[avehtari]

ppc_loo_intervals and ppc_loo_ribbon have optional psis_object argument, but other ppc_loo_* functions don't. It would be nice if all of them would accept psis_object argument

[jgabry]

Yeah I agree

[avehtari]

I can fix this, but to not break anything I assume the other ppc_loo_* functions need to still keep lw argument. What should happen if both are given?

[jgabry]

Thanks. Is there any reason to supply both? If not then maybe an error saying only one should be supplied.

[jgabry]

Actually I'm working on the next bayesplot release now (to make sure we stay compatible with ggplot2, thanks for tagging me on that). So I can fix this issue too.

What's strange is that I just noticed ppc_loo_ribbon has the lw argument but doesn't use it for anything. For ppc_loo_intervals and ppc_loo_ribbon it should be either intervals or psis_object, there's no need for lw. Anyway, I'll fix that along with adding the optional psis_object argument to the other functions.

[jgabry]

And actually if both are given I guess we don't need to error, we can just use lw since that will skip the step of extracting the weights from the psis_object.

I'll work on this now.

Edit: see PR #312