feat!: per-language [[headers]] rules with existingStrategy

[hubcio]

Replace the style-keyed [mapping.STYLE] config and useDefaultMapping
with a per-language [[headers]] rule list carrying styles (preferred
first) and an existingStrategy (replace|skip|error). Fixes #210:
format no longer stacks a duplicate header when a file already has a
header in a different comment style.

An existing header is detected style-agnostically by a whole-word,
case-insensitive scan for the configured keywords. A foreign-but-listed
header (e.g. /* */ under a // rule) is migrated to the preferred style,
not duplicated. One in an unlisted style, or a stray keyword that cannot
be normalized, is left unchanged and reported as "foreign": under
existingStrategy=replace this fails the run (exit non-zero) rather than
silently passing un-normalized; skip leaves it and exits zero. check
adds a "foreign" category; --output JSON gains foreign/skipped/conflict.

format and remove now write files atomically (temp, fsync, rename);
a symlinked path is replaced by a regular file, and source permissions
are preserved on a best-effort basis.

BREAKING CHANGE: [mapping.STYLE] blocks and useDefaultMapping are
removed. Replace each [mapping.FOO] { extensions = [...] } with a
[[headers]] rule carrying styles = ["FOO"], and rename
useDefaultMapping to useDefaultHeaders. A 6.x config is rejected
with a migration hint. format now exits non-zero when it finds an
existing license-looking header it cannot normalize to the preferred
style; hawkeye remove's --output "removed" entries are now bare paths
(previously path=removed).

Tested on downstream hawkeye users by migrating their configs:
apache/opendal, datafuselabs/databend and apache/fory match 6.5.1
with no new diagnostics; apache/iggy needs one gitignored file excluded.

Closes #210

[tisonkun]

Thanks for your contribution @hubcio! I'm going to review this patch on the weekend.

Feel free to ping me next week if I miss this review.

[tisonkun]

Is it? Untracked files that do not ignored should be processed always.

[tisonkun]

This is a huge PR I just noticed.

Let me review it in the following days.

Actually I'm considering a full rewrite to cover:

1. Use snake_case options always
2. Drop GitHub Actions support; everyone can use cargo-binstall and run the command as a faster way.
3. Rework the integration tests framework.
4. Forward quite a few Rust best practices.

That said, I'm taking these on my own. And thanks a lot for @hubcio your patch and nudge to me for finally working on these.

Hopefully I can get all these arranged and delievered in July.

[hubcio]

Thanks @tisonkun. Let me know if you need any help.