I read though this as thoroughly as I could, although of course with this sort of systematic updating it's easy to get lost in the patterns and miss details, so I tried to look particularly carefully at the test cases just to see if they agreed with what I'd expect the new behavior to be, and they do seem to.

My sense is that it's better to get this in earlier rather than latter so that it has some time to soak in at least one alpha and then the betas, since then there's more chance of us or someone else exercising these things indirectly as part of testing something else.

Noting that we still have some commands other than terraform apply which take a plan file as a positional argument, I had a further idea here although I'm just noting it for completeness and not intending to try to increase the scope of this PR:

While terraform apply tfplan has nice ergonomics because the plan file is obviously the "direct object" of the "apply" verb, the positional form for other commands has always felt less clear to me. Given that we're already making some breaking changes here anyway, perhaps a further PR could go further by replacing those plan file position arguments with named arguments instead, like this:

terraform graph -plan=tfplan

This would avoid relying on the plan file having a meaningful name in order for the meaning of that argument to be clear, and also just generally standardize that we typically use named arguments because they are better for allowing backward-compatible additions in future. (with terraform apply tfplan of course remaining one special exception, justified due to its prominence in the CLI workflow.)