fix: disambiguate archive entries when multiple inputs share the same basename

[toller892]

Problem

When compressing multiple directories with the same basename (e.g., /mnt/d/Temp/GOG/Scripts and /mnt/d/Temp/kodi/Scripts), ouch strips paths to just the basename using file_name(). This causes:

• Zip: "Invalid zip archive - Duplicate filename: Scripts/"
• Tar: both directories merge into one, causing data loss on decompress

Fixes #978

Root Cause

In all three archive builders (tar, zip, 7z), the build_archive function:

1. Calls cd_into_same_dir_as(explicit_path) to change cwd to the parent directory
2. Uses explicit_path.file_name().unwrap() to get just the basename
3. The walker walks from the basename, so archive entries are just Scripts/...

When two inputs share the same basename, they collide.

Fix

When input paths have duplicate basenames, compute the common ancestor directory of all inputs and store archive entries relative to it, including enough parent path components to disambiguate (e.g., parent1/Scripts vs parent2/Scripts).

When basenames don't collide, behavior is unchanged (backward compatible).

Example

# Before (broken):
ouch compress /a/parent1/Scripts /a/parent2/Scripts archive.tar
# Archive contains: Scripts/a.txt, Scripts/b.txt  <-- COLLISION!

# After (fixed):  
ouch compress /a/parent1/Scripts /a/parent2/Scripts archive.tar
# Archive contains: parent1/Scripts/a.txt, parent2/Scripts/b.txt  <-- no collision

Files Changed

• src/utils/fs.rs - Added has_duplicate_basenames() and common_ancestor() helper functions
• src/archive/tar.rs - Use common ancestor approach when basenames collide
• src/archive/zip.rs - Same fix for zip builder
• src/archive/sevenz.rs - Same fix for 7z builder
• tests/integration.rs - Added tests for tar and zip with duplicate basenames

Test Results

All 91 tests pass (34 unit + 39 integration + 3 mime + 13 UI + 2 utils), including the 2 new tests:

• compress_dirs_with_same_basename_tar
• compress_dirs_with_same_basename_zip

[Mazwak]

My 2 cents.

I would behave like tar/zip by default: Use path from command line, stripping leading /.

And maybe add something like --shorten-path.

[marcospb19]

I would behave like tar/zip by default: Use path from command line, stripping leading /.

I'm slightly more favorable to this suggestion than the biggest common ancestor approach used in this PR.

Mainly due to predictability.

@tommady @valoq can I get u guys opinions?

(If we change the default behavior we wouldn't need a flag too.)

[tommady]

I would behave like tar/zip by default: Use path from command line, stripping leading /.

I'm slightly more favorable to this suggestion than the biggest common ancestor approach used in this PR.

Mainly due to predictability.

@tommady @valoq can I get u guys opinions?

(If we change the default behavior we wouldn't need a flag too.)

As I mentioned in the issue comment below, I will follow your advice.

#978 (comment)

[valoq]

The tar and zip approach is better, yes.

A few more potential issues:

• Cross-drive inputs (e.g. C:\scripts and E:\Backup\Scripts) share no component, so common_ancestor returns an empty path and set_current_dir("") fails on Windows.
• Passing the same path twice now fails. The shared folder it finds is the whole path, so there is nothing left below it to add to the archive, and it errors. Before this change, the same path twice just produced a duplicate entry.
• When inputs share only the root, entries keep their full path from the root instead of only the part needed to tell the files apart
• The 7z change needs unit tests for all cases