fix(plugin): decode diff Git paths as UTF-8

[sylvesterkaczmarek]

Summary

Decode Git filename lists explicitly as UTF-8 in diff ranking instead of using Python's process-locale decoder.

Fixes #533.

Reproduction / evidence

Current upstream main at 37bf87a692fc72d41f7312cc48808d699d204fba has two direct text-mode Git subprocesses in generate_rank_input.py without an explicit encoding:

• git diff --name-status -z for committed, staged, and unstaged changed paths;
• git ls-files --others --exclude-standard -z for working-tree untracked paths.

With text=True, Python chooses the process locale encoding. On Windows with a non-UTF-8 ANSI code page, Git's UTF-8 filename output can therefore decode incorrectly or fail before ranking begins. The repository path itself can be ASCII; a changed filename containing non-ASCII characters is sufficient.

This exact call site was independently identified in the earlier closed #293 investigation while addressing #192. That PR documented failures for non-ASCII Git filenames and explicitly noted this helper as a second call site. The current helper still contained both locale-dependent subprocesses.

The regression in this PR instruments subprocess.run inside the actual Python helper and refuses every Git text invocation unless encoding == "utf-8". It exercises both the changed-path and untracked-path flows using synthetic 測試.py and 新規.py names. On the unfixed helper the assertion fails because encoding is absent.

Root cause

The diff-ranking helper launches Git directly rather than routing these path-list operations through the already-hardened workbench Git helper. Its text-mode subprocesses therefore inherited Python's locale decoder.

Fix

Add encoding="utf-8" to both text-mode Git filename-list subprocesses:

• changed-path enumeration;
• untracked-file enumeration.

No error handler or replacement decoding is added. Invalid UTF-8 still fails rather than silently changing a filename.

Tests / validation

Added diff-rank-git-encoding.test.ts. The test imports the real bundled Python helper, replaces only its subprocess boundary, and verifies:

• every text-mode Git call requests UTF-8;
• committed non-ASCII paths are parsed correctly;
• working-tree changed and untracked non-ASCII paths are preserved;
• synthetic path output is normalized for cross-platform CI.

The branch is based directly on current upstream main at 37bf87a692fc72d41f7312cc48808d699d204fba and is not behind it. Production change: exactly 2 additions, 0 deletions. The remainder is focused regression coverage.

Full repository tests cannot be run in this execution environment because the repository cannot be cloned here. Pushed-head CI remains the authoritative full-suite validation.

Risk

Low. Git filename decoding becomes deterministic UTF-8 instead of locale-dependent. ASCII paths are unchanged, and binary Git operations are untouched.