Add voice-assistant support & TUI demo

[graniet]

Adds first-class audio handling to the LLM stack and a runnable voice-assistant example:

• New MessageType::Audio with helpers (has_audio, audio_data) and a cond!(has_audio) macro.
• AgentBuilder / ChatWithMemory now accept an optional STT provider and launch an async transcription pipeline using Whisper.
• Back-end updates (OpenAI, Anthropic, Google, Azure) gracefully ignore audio when building API payloads.
• Memory tweaks: audio messages are filtered out of sliding-window recall to keep prompts clean.
• Example agent_audio_example.rs: records PCM via cpal, encodes WAV with hound, shows live conversation in a ratatui/crossterm TUI, and drives a two-agent (transcriber → assistant) workflow.
• Feature-gated dependencies (cpal, hound, ratatui, crossterm, anyhow, parking_lot) under audio-example to avoid bloat for non-audio builds.

[reneleonhardt]

Wow, great work @graniet 🚀
What is missing? Could a CI workflow be added to check for regressions?

[Copilot]

Pull Request Overview

This PR adds voice-assistant support and a terminal UI demo to the LLM stack. Key changes include:

• Introducing a new audio message type with helper methods and a cond!(has_audio) macro.
• Enhancing memory and chat wrappers to filter out and transcribe audio messages by integrating an optional STT provider.
• Adding a runnable TUI-based voice-assistant example along with feature-gated audio dependencies.

Reviewed Changes

Copilot reviewed 13 out of 14 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/memory/sliding_window.rs	Filters out audio messages from the sliding window memory recall.
src/memory/mod.rs & cond_macros.rs	Adds the new HasAudio condition and helper documentation.
src/memory/chat_wrapper.rs	Integrates an STT provider to spawn a transcription pipeline for audio messages.
src/chat/mod.rs	Introduces new methods to work with the Audio messageType for audio processing.
src/builder.rs; src/agent/builder.rs	Updates agent and builder constructions to support optional STT providers.
src/backends/*	Updates backend implementations to gracefully ignore audio messages.
examples/agent_audio_example.rs	Provides a complete demo integrating recording, transcription, audio processing, and terminal UI.
Cargo.toml	Adds optional dependencies for audio processing and the TUI example.

Comments suppressed due to low confidence (1)

src/memory/sliding_window.rs:182

• Consider adding a comment to explain why audio messages are filtered out in the sliding window recall, to clarify the intent for future maintainers.

        messages.retain(|m| !m.has_audio());

[Copilot]

Consider replacing eprintln with a proper logging framework to ensure consistent error handling across production code.

[Copilot]

Consider handling errors explicitly when writing WAV samples instead of using unwrap to prevent potential panics during audio processing.

Suggested change

	writer.write_sample(s).unwrap();
	}
	writer.finalize().unwrap();
	if let Err(e) = writer.write_sample(s) {
	eprintln!("Error writing WAV sample: {e}");
	return Vec::new(); // Return an empty buffer on error
	}
	}
	if let Err(e) = writer.finalize() {
	eprintln!("Error finalizing WAV file: {e}");
	return Vec::new(); // Return an empty buffer on error
	}