v3.0.0

This release is a major foundation update for Rapida Voice AI. It adds workflow orchestration, external recording storage, stronger telephony support, and cleaner runtime observability across audio and LLM pipelines.

  ### Highlights

  - Added **Agentflow foundation** for node-based voice-agent workflow orchestration.
  - Added experimental **AgentKit** execution support with improved gRPC options and event handling.
  - Added **external recording upload** support so assistant recordings can be pushed to customer-managed cloud storage.
  - Added **assistant configuration APIs** and deployment REST APIs.
  - Added **Telnyx telephony provider**.
  - Strengthened **SIP lifecycle**, inbound route middleware, call status metrics, and shared call status constants.
  - Standardized observability across audio, LLM, webhook, storage, and telephony pipelines.
  - Improved recording reliability by persisting recordings before session finalization.
  - Added packet-controlled input gate and better early-audio handling.
  - Added debugger deployment API.
  - Upgraded Go runtime and security dependencies.

  ### Fixes and Improvements

  - Fixed base URL handling for invites.
  - Fixed user role invite validation.
  - Fixed webhook timeout bounds, webhook ID formatting, and call data normalization.
  - Fixed ambient media option propagation.
  - Fixed model stream connection ownership.
  - Fixed AgentKit executor close-state handling.
  - Improved RN noise chunk resampling.
  - Added YAML-based configuration for clearer setup.

  ### New Contributor

  Thanks to @a692570 for adding Telnyx telephony provider support.

  **Full Changelog**: https://github.com/rapidaai/voice-ai/compare/v2.3.0...v3.0.0