`air format <file>.R` and `air format --stdin-file-path <file>.R` should respect `exclude` and `default-exclude`

[DavisVaughan]

Extracted from #471

Currently if a user does

air format <file>.R

then <file>.R is unconditionally formatted, full stop, regardless of exclude and default-exclude.

This means that air format renv/activate.R would try to format renv/activate.R, even though it is part of default-exclude and should never be formatted. Similar results with any user provided excludes.

I think this is wrong!

In particular, it makes it more difficult for tools such as pre-commit and RStudio to invoke air. Those tools have no way of knowing what excludes or default-excludes are, so when a user edits or saves renv/activate.R, both of them are very likely to just call:

air format renv/activate.R

This was a problem for ruff as well, so much so that they added a special --force-exclude option that they use in their default pre-commit hook:
https://github.com/astral-sh/ruff-pre-commit/blob/1629a02ed57c23345951c75547bfe5fb73970fd3/.pre-commit-hooks.yaml#L15

For them, --force-exclude means "Respect excludes even for files provided directly on the command line".

I'd like to make the argument that we should just always respect exclude and default-exclude for files provided directly on the command line.

It seems very unlikely that a user would really want to ever run air format renv/activate.R, but there is a real problem with the fact that pre-commit (and eventually RStudio when it uses stdin) do this right now and will change these files. So I'd advocate for the safer default.

[DavisVaughan]

Similarly, if you have exclude = ["exclude/"] and you do air format exclude/subdir/ then we currently will format everything inside of subdir/! This is also wrong.

I think in general we should climb the ancestors of every explicitly provided path to see if it should be excluded before recursing into it (and once we start recursing into it, we no longer need to look at any ancestors)