Optimize comparisons for selected files

[EdwardBlair]

Summary

• avoid constructing a complete snapshot-tree comparison when undochange, diff, or xadiff is given explicit paths
• compare only the selected paths in both library and D-Bus modes while preserving status, ignore-filter, and undo behavior
• perform D-Bus comparisons inside snapperd, with compatibility fallback for older daemons
• reject paths outside the configured subvolume and balance comparison/mount cleanup on success and failure

Fixes #251.

Testing

• make check in a disposable openSUSE Tumbleweed Podman container
• focused selected-files.test coverage for modified, created, deleted, type-changed, permission-changed, ignored, duplicate, and rejected paths, plus undo execution
• testsuite-real: all 16 tests passed against a loopback Btrfs filesystem in a privileged disposable Podman container, including permissions, ownership, error handling, xattrs, and ACL restoration
• end-to-end A/B benchmark comparing the parent commit with this branch through the real snapper CLI, system D-Bus, snapperd, and a loopback Btrfs filesystem:
  • 100,000 baseline files
  • 20,000 unrelated endpoint differences (10,000 modified, 5,000 deleted, 5,000 created)
  • one selected file restored and correctness-checked per run
  • one warm-up followed by seven timed runs per version
  • parent median: 303.0 ms
  • optimized median: 11.3 ms
  • end-to-end speedup: 26.78x
• manual end-to-end A/B test on an Arch Linux Btrfs /home configuration with real snapshot history:
  • snapper diff 1732..0 ~/.codex/config.toml: 36.355 s with Arch's snapper 0.13.1 / libsnapper 8.0.0, 0.011 s with this branch (~3,300x faster)
  • snapper undochange 1732..0 ~/.codex/config.toml: 21.235 s with Arch's snapper 0.13.1 / libsnapper 8.0.0, 0.011 s with this branch (~1,930x faster)
  • the undo operation reported create:0 modify:1 delete:0, confirming that it performed a real restoration
  • timings include the CLI, D-Bus, snapperd, comparison, and filesystem work

I use Arch btw 😉

[EdwardBlair]

Optional automated CLI/D-Bus regression coverage is available as a stacked draft PR against this PR’s head branch: EdwardBlair#2

It runs the real snapper CLI and snapperd over an isolated D-Bus with loopback Btrfs, verifies that diff and undochange use GetFilesForPaths without falling back to CreateComparison, and checks that only the selected file is restored. I have kept it separate for now so it can be reviewed or included without expanding this PR unconditionally.

[aschnell]

What I dislike about the PR is that need features are added to snapperd. In the end I cannot see a reason why this cannot be entirely be implemented in the client (with help of libsnapper).

[EdwardBlair]

I see what you mean now this is the wrong layer. If I understand correctly the client should use the existing snapshot mount paths and libsnapper to compare only the explicitly requested files.. I hadn't realised this was possible because the surface-level/proxy API does not obviously expose the underlying snapshot path, which made it look as though the comparison had to stay behind the daemon boundary. I shall investigate this. It looks like it will be a much leaner change!