feat: add conversation recorder extension

[Roei-Bracha]

Conversation Recorder Extension

This PR adds the conversation_recorder extension, which allows for recording both user and agent audio during a session, mixing them into a single WAV file, and saving them to various storage backends.

Changes Overview

The PR introduces a new extension located at ai_agents/agents/ten_packages/extension/conversation_recorder/. Below is a breakdown of the included files:

File	Description
extension.py	Core extension logic. Manages the recording lifecycle, listens for on_user_joined/on_user_left events, and coordinates audio mixing and storage.
audio_mixer.py	Handles mixing multiple PCM audio streams into one. Includes support for real-time resampling (e.g., matching 16kHz user audio with 24kHz agent audio).
storage.py	Provides an abstraction for storage backends. Supports Local Filesystem, Google Cloud Storage (GCS), and S3-compatible storage.
README.md	Comprehensive documentation on configuration, features, and graph integration.
manifest.json	Extension metadata and definition of all configuration properties.
requirements.txt	Lists dependencies: numpy, scipy, google-cloud-storage, and boto3.
addon.py / __init__.py	Necessary boilerplate for TEN Agent extension registration and loading.
property.json	Default values for the extension's properties.

Key Features

• Multi-Source Audio Mixing: Successfully combines user and agent audio frames into a single high-quality WAV recording.
• Graceful Shutdown: Implements signal handlers (SIGTERM, SIGINT) and atexit hooks to ensure recordings are correctly flushed and closed even if the agent process is interrupted.
• Pluggable Storage: Native support for uploading recordings directly to GCS or S3 buckets, or saving them locally.
• Event-Driven Recording: Can be configured to start automatically on agent startup or be triggered by specific commands like on_user_joined.

How to use

Integrate the extension into your graph by:

1. Connecting the pcm_frame outputs from audio sources (e.g., streamid_adapter and v2v) to the conversation_recorder node.
2. Connecting the on_user_joined and on_user_left commands from your RTC extension to the recorder.
3. Configuring the storage_type and relevant credentials in the graph properties.