Verify that formatters comply with the treefmt Formatter Specification

[jfly]

See this recent PR: #356 (adding a wrapper for a formatter that doesn't accept multiple files). See our spec.

We could also confirm that formatters do not write to files when nothing has changed. See our spec.

I believe the other rules of the spec are hard/impossible to confirm harder to confirm.

[zimbatm]

A good way to test this would be if we have two sample: one formatted, and one unformatted. That should be enough to test most of the spec's feature in a nix build sandbox.

[jfly]

Or perhaps one unformatted sample that we format twice?

[zimbatm]

It's something that has been on my mind for a while. It would be nice if treefmt itself could detect it, and then emit warning like "formatter XYZ violates rule nr3"