Avoid clone in fallback

[yanns]

The first commit creates test coverage on state clone in fallback.
The second commit removes one clone.

This is an extraction of a part of #2865

[jplatte]

Nice 👍🏼

[yanns]

actually, we don't need a Mutex as we're using atomic values: d9ef8e0

[yanns]

@jplatte do we prefer "squash and merge"?
Or I could make a better git log with only 2 commits, and use "rebase and merge"?

[jplatte]

I'd say it doesn't matter much here. I implied in the other PR that squash doesn't preserve authorship, but actually I think GitHub will add Co-authored-by lines for separate authors in squash-merge commits.

Generally I default to squash-merge in axum bc. that's what David pretty much always used, making some exceptions for larger PRs with well-separated commits.

[yanns]

OK thanks. I'll keep the history here to show that the 2nd commit decreases the number of clones.

[jplatte]

One nitpick about the commit messages, please capitalize them in the future (unless there's some prefix like macros: , though I haven't seen that used much in this repo anyways).