fix(scan): index files under non-ASCII (CJK) directory names

[Arvin-Hugh]

Problem

Source files under a directory whose name contains non-ASCII characters (e.g. CJK / Chinese folder names) are silently skipped during indexing — no error, no warning. They are not gitignored and parse fine; they simply never enter the index.

On a real .NET solution that groups code into Chinese business-domain folders, only 114 of 902 .cs files were indexed — every skipped file lived under a directory with a non-ASCII name; every indexed file lived under an all-ASCII path.

Minimal reproduction

REPRO=/tmp/cg-cjk-repro
mkdir -p "$REPRO/src/english" "$REPRO/src/中文目录"
printf 'public class Foo {}\n' > "$REPRO/src/english/Foo.cs"
printf 'public class Bar {}\n' > "$REPRO/src/中文目录/Bar.cs"
cd "$REPRO" && git init -q && git add -A && git commit -qm init
codegraph init "$REPRO" && codegraph index "$REPRO" --force
codegraph files --format flat   # only src/english/Foo.cs shows up

Root cause

collectGitFiles and getGitChangedFiles read git ls-files / git status output verbatim. git defaults to core.quotePath=true, which octal-escapes and double-quotes any path containing a byte outside ASCII. So a tracked file at src/中文目录/Bar.cs arrives as:

"src/\344\270\255\346\226\207\347\233\256\345\275\225/Bar.cs"

The surrounding quotes and escapes mean isSourceFile() sees an extension of .cs" (trailing quote), which isn't in EXTENSION_MAP, so the file is dropped. All-ASCII paths are emitted unquoted and are unaffected — which is exactly why only English-named folders were indexed.

Fix

Run the three path-reading git commands with -c core.quotePath=false so non-ASCII paths come through as real UTF-8:

• git ls-files -c --recurse-submodules (tracked)
• git ls-files -o --exclude-standard (untracked)
• git status --porcelain --no-renames (incremental sync)

Testing

• Added a regression test (Non-ASCII (CJK) paths) that commits one file and leaves one untracked under a CJK-named directory, plus an ASCII control, and asserts all three .cs files are scanned with no quoting artifacts in the returned paths. It fails before the change (1 of 3 found) and passes after.
• Full vitest run shows no new failures attributable to this change.
• Verified end-to-end on the real repo: indexed file count went from 114 → 892, and the previously-missing Chinese-path source is now fully present.

No existing issue filed for this; happy to open one to track if preferred.