Public-repo polish: README, INSTALL.md, install.sh, doctor, env file, CI

[rodaddy]

What

• Rewritten README (three doors, line-choice cascade, config table, commands).
• INSTALL.md: five-step walkthrough incl. standing up scorbo2's server_omnivoice.py (port 8181), cutting a real voice with ffmpeg, writer-model table (Ollama / LM Studio / llama.cpp / oMLX / hosted), troubleshooting table.
• install.sh: POSIX sh; finds Python 3.11+, venv via uv or stdlib, installs, writes agent-voice.env from the example, --demo voice on macOS, --dev tools, ends with doctor.
• agent-voice doctor: python / player / tts / voices (incl. ref.wav length + PCM check) / writer (/v1/models) / ffmpeg, each failure with a fix: line; exit 1 on required failures.
• Config.from_env reads agent-voice.env from cwd (environment wins); default TTS URL is now http://127.0.0.1:8181 to match scorbo2's README.
• CI: ruff / format / mypy / pytest on ubuntu+macos x 3.11-3.13, plus an install.sh job.
• CONTRIBUTING, CHANGELOG, agent-voice.env.example.

Verified (RUNNING, 2026-08-27)

• ruff check, ruff format --check, mypy --strict, pytest (21 passed) on Python 3.12.
• sh install.sh --demo from a clean clone in a scratch dir: venv + install + env file + demo voice + doctor report (one FAIL, TTS not configured, with the fix named).
• Same clone, env file pointed at a live OmniVoice server: doctor green for tts, agent-voice speak --voice demo produced a WAV.
• agent-voice doctor all-green on the dev box against a remote OmniVoice + local oMLX writer.

CI on this PR is the first run of the workflow (PROPOSED until green).

🤖 Generated with Claude Code

https://claude.ai/code/session_01FggfyFMVwmG32Hpx4EAoeQ