Pin UTF-8 encoding for cache and manifest I/O

[azizur100389]

Summary

On Windows the default text codec is cp1252, so Path.read_text() / Path.write_text() raise UnicodeDecodeError / UnicodeEncodeError whenever the graphify cache or manifest holds non-ASCII content — for example:

• Chinese / Japanese / Korean identifier names extracted from source code
• File paths under directories with accented Latin or CJK characters
• Emoji or non-ASCII labels in semantic extraction output

Four call sites are affected:

• cache.load_cached() — read_text(encoding="utf-8") + catch UnicodeDecodeError
• cache.save_cached() — write_text(encoding="utf-8") with ensure_ascii=False so CJK/emoji stay readable UTF-8 instead of \uXXXX escapes
• detect.load_manifest() — read_text(encoding="utf-8")
• detect.save_manifest() — write_text(encoding="utf-8") with ensure_ascii=False

The repo already uses encoding="utf-8" in __main__.py and export.py; this extends the same convention to the file types most likely to hold user data.

Test plan

• New tests/test_encoding.py (13 tests) covers cache + manifest roundtrips with CJK, Japanese, Korean, accented Latin, emoji, and mixed-script payloads — including non-ASCII filenames
• Verifies cache files on disk are valid UTF-8 JSON readable by external tools
• No regressions in existing tests

[azizur100389]

Closing as redundant — checked the v6 default branch and graphify/cache.py already uses encoding="utf-8" (line 87, 95), and detect.load_manifest/save_manifest already use it too. This PR was opened against v3 before realizing v6 was the active branch. Sorry for the noise.