`power_size_results()` changes for `power_pool()` printing

[fredjaya]

@AngusMcLure following on from #36 in a PR so it's easier to provide in-line comments before incorporating into the feature branch.

Changes:

• Add more attributes to power_size_result class
• Group related attributes in lists (e.g. access to results$power now results$stat_test$power)
• Update print method to output a summary line at the end, instead of marking the inferred param
• Update roxygen docs to fix undocumented params, left descriptions as .

Additions:

• test for constructing power_size_result

[AngusMcLure]

A couple comments that don't prevent the merge:

• sample_size_pool_random doesn't have the new output type implemented yet?
• sample_size_pool returns the input power in the output. Strictly speaking, the function identifies the smallest number of clusters that achieves at least this much statistical power, so the power of the identified design will often be more than this (though perhaps not by much). I think it might be worth returning the actual survey power instead or in addition i.e. feeding the sample size back into the powerpool* function to get the actual power. Happy to discuss when chatting in person

[fredjaya]

Just pushed for sample_size_pool_random. Let's discuss tomorrow!

[AngusMcLure]

Great. Also, just noting a couple related thoughts so I don't forget.

Feeding design back into powerpool is probably slow and inefficient since it recomputes the fisher information.

However, the powerpool functions have a lot of shared code across them, with the one thing being different is how the fisher information are calculated. So I think it might make sense to have a backend power_ function that takes in the fisher information instead of the sample design.

This will reduce code duplication, avoid re-computation of fisher information within samplesize and make it easier to extend to variants of power calculations (e.g. power based on precision of estimates, rather than comparing to a threshold, or power when comparing two samples rather than one sample to a threshold)