mv: create the cross-device destination exclusively and chown it by fd

[krosci]

rename_file_fallback kept the destination descriptor across the copy but preserve_ownership re-resolved the destination by path and called lchown, so a concurrent path-swap could redirect the chown to another inode. The comment above the block claimed a guarantee the code did not provide.

The cross-device leaf now chowns through the descriptor it already holds with fchown, and the destination is created with O_EXCL, matching GNU. create_dest_restrictive gains an exclusive flag, while cp keeps truncating a pre-existing destination.

Fixes #14599.

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/date/date-locale-hour (passes in this run but fails in the 'main' branch)
Skipping an intermittent issue tests/misc/io-errors (passes in this run but fails in the 'main' branch)
Skipping an intermittent issue tests/tail/inotify-dir-recreate (passes in this run but fails in the 'main' branch)
Skipping an intermittent issue tests/tail/pid-pipe (passes in this run but fails in the 'main' branch)
Skipping an intermittent issue tests/tail/pipe-f (passes in this run but fails in the 'main' branch)
Congrats! The gnu test tests/cp/cp-mv-enotsup-xattr is no longer failing!
Congrats! The gnu test tests/csplit/csplit is no longer failing!
Congrats! The gnu test tests/date/date-tz is no longer failing!
Congrats! The gnu test tests/dd/nocache is no longer failing!
Congrats! The gnu test tests/df/df-output is no longer failing!
Congrats! The gnu test tests/du/max-depth is no longer failing!
Congrats! The gnu test tests/env/env is no longer failing!
Congrats! The gnu test tests/fold/fold is no longer failing!
Congrats! The gnu test tests/id/setgid is no longer failing!
Congrats! The gnu test tests/join/join is no longer failing!
Congrats! The gnu test tests/ls/ls-time is no longer failing!
Congrats! The gnu test tests/ls/m-option is no longer failing!
Congrats! The gnu test tests/misc/printenv is no longer failing!
Congrats! The gnu test tests/mv/mv-special-1 is no longer failing!
Congrats! The gnu test tests/od/od is no longer failing!
Congrats! The gnu test tests/printf/printf is no longer failing!
Congrats! The gnu test tests/ptx/ptx is no longer failing!
Congrats! The gnu test tests/seq/seq-precision is no longer failing!
Congrats! The gnu test tests/shred/shred-remove is no longer failing!
Congrats! The gnu test tests/shuf/shuf is no longer failing!
Congrats! The gnu test tests/sort/sort is no longer failing!
Congrats! The gnu test tests/split/split-io-err is no longer failing!
Congrats! The gnu test tests/stat/stat-fmt is no longer failing!
Congrats! The gnu test tests/stat/stat-nanoseconds is no longer failing!
Congrats! The gnu test tests/test/test is no longer failing!
Congrats! The gnu test tests/touch/now-owned-by-other is no longer failing!
Congrats! The gnu test tests/wc/wc is no longer failing!
Note: The gnu test tests/csplit/csplit-heap is now being skipped but was previously passing.

[krosci]

@sylvestre could you take a look at it? Thanks!

[codspeed]

Merging this PR will not alter performance

✅ 377 untouched benchmarks
⏩ 54 skipped benchmarks¹

---

_{Comparing krosci:mv-fchown-cross-device (80d3d5b) with main (5eb255f)}

Footnotes

1. 54 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩