Add local-first TTS/STT providers, event-driven architecture, and uv/justfile tooling

[lancekrogers]

Summary

Major overhaul that makes Samantha fully usable without any API keys or cloud services. Adds a pluggable provider system for both TTS and STT, replaces the tightly-coupled conversation loop with an event-driven architecture, and modernizes the build/dev tooling.

Local-First Audio Providers

• Kokoro TTS (default) — high-quality local ONNX text-to-speech, no API key needed
• Whisper STT (default) — local speech-to-text via faster-whisper, no internet required
• Edge TTS — free cloud alternative (Microsoft), no API key
• Existing Fish Audio TTS preserved as an optional paid provider for custom voice clones
• Google STT preserved as an optional cloud fallback
• All providers follow a common base class (TTSProvider / STTProvider) for easy extensibility

Event-Driven Architecture

• New EventBus (samantha/events.py) decouples the conversation engine from the UI
• New ConversationEngine (samantha/engine.py) manages the listen → think → speak loop independently of display concerns
• UI subscribes to engine events (listening, thinking, speaking, error, etc.) instead of being called directly
• Adds real-time status callbacks for STT listening/transcribing phases
• Live microphone level visualization during listening state

New CLI Commands

• samantha test — end-to-end mic + speaker test with per-provider diagnostics
• samantha voices — list available TTS voices with locale/gender filtering
• samantha providers — show installed/active TTS and STT providers with install hints

Build & Tooling

• Migrated from setuptools to hatchling build backend
• Added uv as the package manager with optional dependency groups ([whisper], [fish], [edge], [local], [cloud], [all])
• Added python-dotenv for .env file support
• Full justfile system with modular recipes: voice.just, dev.just, install.just
• Recipes for switching providers, going fully local, testing audio, and more

Other Improvements

• Animated microphone waveform indicator for listening state
• Conversation history and config management unchanged (backward compatible)
• Updated CLAUDE.md to reflect new defaults and zero-API-key setup

Changed Files

• samantha/tts/ — New TTS provider package (kokoro, edge, fish + base class)
• samantha/stt/ — New STT provider package (whisper, google + base class)
• samantha/engine.py — New conversation engine (decoupled from UI)
• samantha/events.py — New event bus for engine↔UI communication
• samantha/voice.py — Refactored to provider-agnostic orchestration
• samantha/cli.py — New subcommands (test, voices, providers)
• samantha/config.py — New provider settings and defaults
• samantha/ui.py — Event-driven status display with waveform animation
• pyproject.toml — hatchling backend, optional deps, dotenv
• justfile + .justfiles/ — Full modular just recipe system
• uv.lock — Lockfile for reproducible installs

Test Plan

• samantha --text works with default Kokoro TTS (no API keys)
• samantha works with default Whisper STT + Kokoro TTS (fully local)
• samantha test correctly reports mic and speaker status
• samantha providers shows installed vs missing providers
• samantha voices lists available voices for the active TTS provider
• Switching providers via samantha config tts_provider edge works
• just recipes work (just talk, just text, just voice go-local)
• Fish Audio TTS still works when API key is provided
• Google STT still works as a fallback