fix(media): read ElevenLabs input as UTF-8 - #5469

Reviewed at ca4dc947. Approving. Thanks for the fix!

This was the only file read that needed it. In ELEVENLABS_PY, open(sys.argv[1]) was the one Python read of the text file. Node writes that file with writeFileSync(p, text), which is UTF-8, so on Windows Python's default cp1252 decode garbled anything non-ASCII. Now that read says encoding="utf-8". I also checked the other places that hand text to a subprocess:

• kokoro: hyperframes tts <file> reads the file in Node with "utf-8" and passes the text to Python in argv. Windows hands that to Python as Unicode, so no locale decode happens.
• lyria and musicgen (bgm.mjs): the prompt goes in argv or inside the -c source, with no file read. gemini-auth.py already opens its file with encoding="utf-8" and reconfigures its streams. gemini-tts.mjs doesn't start Python at all.

One thing to do before this can merge: main has moved since this branch was cut, and #5477 changed tts.mjs. tts.mjs still merges cleanly, but skills-manifest.json now conflicts. Could you merge main and regenerate the manifest with bun packages/cli/scripts/gen-skills-manifest.ts? The new commit will need one more quick approval, and I'll take care of that.

I didn't run ElevenLabs or test on Windows.

— Rames