Ensure consistency in the way that no-correlation cases is treated across functions

[AngusMcLure]

Currently there is a standard and a _cluster version of most functions, but not for design effect. This currently relies on fi_pool_cluster to calculate the case where correlation = 0, for the non-cluster case. Both of these might be fine, but fi_pool_cluster currently returns the fisher information for a simple random sample when correlation is set to 0. This might not actually be quite right in some cases as the fisher information in the limit as $\rho \to 0$, when $\rho$ is unknown does not always appear to be equal to the fisher information for simple random sampling. It may be better to throw an error for correlation = 0 in fi_pool_cluster (directing people to use fi_pool instead) and design_effect (directing people to use correlation = NA) instead? I need to think on this more before we action though.