fix(cp): don't delete dest before same-file check with --remove-destination

[koopatroopa787]

Fixes #13144 (GHSA-9p9p-xh9v-6fvx).

Bug

$ echo hello > a
$ cp --remove-destination a ./a
cp: cannot stat 'a': No such file or directory
$ cat a
cat: a: No such file or directory      # file is gone

cp --remove-destination a ./a deletes the file before discovering that
source and destination refer to the same data, destroying it. GNU cp
refuses with cp: 'a' and './a' are the same file and leaves the file
intact.

Root cause

In copy_file, the block that removes dest ahead of time to let
--remove-destination recreate hard-linked destinations used source != dest (a raw Path comparison) as the guard against deleting the source
itself:

if are_hardlinks_to_same_file(source, dest)
    && source != dest
    && matches!(options.overwrite, OverwriteMode::Clobber(ClobberMode::RemoveDestination))
{
    fs::remove_file(dest)?;
}

"a" and "./a" are different Path values but the exact same
directory entry, so the guard doesn't catch this case and the file gets
unlinked before the later same-file check (in handle_existing_dest)
ever runs.

Fix

Compare canonicalized paths instead of the raw Paths. This still allows
deletion for the legitimate case this block exists for — two distinct
directory entries that are hard-linked to each other (e.g. cp --remove-destination file hardlink, covered by
test_remove_destination_with_destination_being_a_hardlink_to_source) —
but skips it when source and dest are actually the same entry, letting
the existing "same file" error fire instead.

Testing

Added test_remove_destination_same_path_does_not_destroy_file
(cross-platform) reproducing the exact GHSA repro. Ran the full cp test
suite (159 tests) with no regressions, plus cargo clippy -D warnings
and cargo fmt --check.

[github-actions]

GNU testsuite comparison:

GNU test failed: tests/cp/sparse-to-pipe. tests/cp/sparse-to-pipe is passing on 'main'. Maybe you have to rebase?
Skipping an intermittent issue tests/date/date-locale-hour (passes in this run but fails in the 'main' branch)
Note: The gnu test tests/seq/seq-epipe is now being skipped but was previously passing.
Note: The gnu test tests/tail/tail-n0f is now being skipped but was previously passing.