DOC: Guidance on when to use filter_parameterization or not

[pllim]

This is a follow-up of #257 (comment) from @eteq

Maybe we want to add an issue ... with a bit more useful guidance on what use cases it's good for? E.g. "approximating filter X with N coefficients yields a curve that is Y% as good as the full data set." It's a minor project to actually do that for a few filters though...