Add tests for behavior of count()

[krlmlr]

None of the existing tests was triggered when the sort order wasn't established.

I see that summarize(.by = ...) has a different default (of returning in the order of appearance), I wonder what the long-term intention for count() is.

[DavisVaughan]

I think it is sort of reasonable to say that the sort order of count() is undefined unless you set sort = TRUE to sort by count size, so we may be able to change it in the future.

See also #6806 where I think the long term data.frame implementation of count/tally would use vec_count() to make it much faster, but that is tough because count() has proven to be fairly complicated.

[DavisVaughan]

😬 I didn't even know this worked and I'm not sure I like it

[krlmlr]

There might be a CRAN package that uses this feature.

[krlmlr]

A random sort order means that count() behaves somewhat differently from summarize(), though. I don't have a strong opinion, we can leave it open for now.

[DavisVaughan]

I'm happy to merge this one as is, I do agree we should at least test it so if we ever change things in the future it forces us to acknowledge that we are breaking something