feat: ksay — Kokoro 82M local TTS replaces say (4x RTF, offline, 9 voices)

- engine_kokoro_tts.py: tts(text, voice, outfile) 0.5s gen 4-5x RTF M2
- ksay: drop-in say replacement, usage: ksay [-v voice] [-o file] "text"
- install-ksay.sh: one-shot setup for macmini (pip + model 310MB + warm)
- voices: af_bella (default), af_heart, am_adam, bf_emma etc
- model ready after: ./install-ksay.sh (pre-downloads hexgrad/Kokoro-82M)
- macmini: git pull gitea main && bash install-ksay.sh && ./ksay "hello"

requirements.txt

@@ -14,3 +14,5 @@ soundfile
 langdetect
 python-dotenv
 openai
+kokoro==0.9.4
+misaki[en]