error in across() when supplied ... without a .fns argument

[simonpcouch]

Closes #6638. cc @DavisVaughan

[DavisVaughan]

@hadley this is the outcome of today's pairing session. Seem reasonable?

In a perfect world we'd mention "... must be empty when .fns isn't supplied" but I still feel like the error is an improvement on what we had before (which was the deprecation warning about supplying ... in across()). The nice part about this error is that check_dots_empty0() will show you which arg caused the problem, and that should be enough for you to figure out you had a typo.

I'm not hugely worried about throwing a great error here anyways. When we escalate the deprecation so that not supplying .fns is an error, we can remove all of this anyways.

We don't need to worry about catching this in the "expansion" path, because expansion always falls back to this "evaluation" path of across() if dots are supplied in any way.