Add fitness exercise finder skill

[anweat]

What changed

• add a Codex skill for searching the exercise dataset by natural-language query, body part, equipment, and target muscle
• add Chinese aliases for common body parts, equipment, muscles, and exercise names
• return multilingual instructions, thumbnails, GIF links, source links, and required Gym visual attribution
• add a standard-library-only Python CLI with local dataset discovery and a bounded upstream download fallback
• add unit coverage for ranking, aliases, malformed input, and media URLs

Why

The dataset already contains rich searchable metadata and multilingual reference material, but agents need a small deterministic workflow to retrieve grounded exercise examples instead of relying on memory.

User impact

Users can invoke $find-fitness-exercises with requests such as “找几个哑铃练胸动作” and receive concise, source-backed examples with Chinese steps and reference media.

Validation

• python -m unittest discover -s skills/find-fitness-exercises/tests -v — 8 tests passed
• ruff check skills/find-fitness-exercises/scripts skills/find-fitness-exercises/tests — passed
• python -X utf8 .../quick_validate.py skills/find-fitness-exercises — skill valid
• real-data searches verified against all 1,324 records
• standalone upstream download fallback verified (17,391,530 bytes; 1,324 records)