Improve test coverage for helper.Err*() functions

I had a bit of a WTF when first reading the error helper code, because we wrap errors on the non-fmt functions, but clobber them on the fmt-functions. I've had a few "why don't you use the helper" comments on my MRs & said something to this effect. But apparently this edge case isn't well-known.

Digging into it further it's a regression in a commit in May 2020, so old that we probably want to keep it for now & not revert.

I had another MR to migrate away from this API in !2859 (closed) whether we do something like that is another matter, this is just a test improvement asserting the (inconsistent) existing behavior.