fix(extraction): handle non-ASCII paths in git fast paths

[thismilktea]

Summary

• switch git-backed file discovery to -z / NUL-delimited parsing
• switch git-backed change detection to git status --porcelain -z
• add regression coverage for CJK directory names in both discovery and sync paths

Problem

Files under non-ASCII directory names could be silently skipped during indexing because git fast paths were parsing
human-readable text output rather than machine-readable path output.

When Git returned quoted/escaped non-ASCII paths, CodeGraph could treat that display-format output as a literal file
path and skip the file before it ever reached extraction.

Fix

This PR updates the git-backed scan/change-detection paths in src/extraction/index.ts to use -z machine-readable
output and parse NUL-delimited records instead of newline-delimited text output.

That avoids Git quoted/escaped path output from being treated as a literal path, which restores correct handling for
files under CJK and other non-ASCII directory names.

Test plan

• npx vitest run __tests__/extraction.test.ts -t "issue #541"
• npx vitest run __tests__/sync.test.ts -t "CJK directories via git"
• npm run build

Close #541