Add SE5 Fix encoding errors plugin

[niksedk]

New SE5 plugin Fix encoding errors (se5/FixEncodingErrors). It repairs UTF-8 text that was read with the wrong code page and saved again (mojibake):

• résumé → résumé, don’t → don’t, 25° → 25°, 😀 → 😀
• rã©sumã© → résumé when the text was lowercased after it broke (a user report: Madam Secretary SRTs extracted from MKV)
• Привет / Привет → Привет, and Greek, Hebrew and Arabic misread as Western
• lost 0x9D (â€ → ”), no-break space turned into a normal space (à → à), BOM as text (), text broken twice

Matroska text tracks are UTF-8 by spec, so SE shows what is stored; the damage happened before muxing.

How it works: each character is mapped back to its byte in the code page and runs that form valid UTF-8 are decoded. Code pages: 1252, 1250, 1251, 1253, 1254 and 1257, with auto-detect.

False-fix guards:

• A decode must land in a plausible script, so real Macedonian ој never becomes Greek μ.
• Ambiguous matches (Não”, irmã…) are fixed only when the subtitle is clearly broken: it has unambiguous matches and they cover ≥30% of its non-ASCII characters.

Checked:

• 43 unit tests.
• False-positive sweep over all SE UI translations and word lists (about 359k lines, 35 languages). With auto-detect the only change is one line that really is broken.
• Round-trip: every language file breaks and fixes back at 100% through its natural code page.

UI follows Italic each line: a code-page picker, All/Selected lines scope, and checkable before/after cards with diff highlighting. Light and dark were checked headless. Not yet run inside a real SE.

After merge: dispatch fix-encoding-errors.yml with release=true, then add it to se5-plugins.json.

🤖 Generated with Claude Code