Add Dockerfiles and Docker examples

[xashr]

Summary

Docker support for audio.cpp is currently limited and focused on development (see Docker.md).

This PR adds Docker support to help users get started quickly. It provides:

• Multi-stage Dockerfiles (CPU and CUDA) with separate build and runtime stages
  • Inspired by llama.cpp (hence the .devops folder naming)
  • "Full" images with all binaries and the multiplexer (similar to llama.cpp's full images)
• Ready-to-run examples for Pocket TTS (CPU & CUDA)
  • CLI: run Pocket TTS via docker run
  • Server: run Pocket TTS via Docker Compose

Tested on Ubuntu Linux with an RTX 5090 and an Intel CPU.

Quick Start

CLI:

1. Download Pocket TTS models into the examples/docker/models/ folder
2. Run from repo root: docker build -f .devops/cuda.Dockerfile -t local/audiocpp:full-cuda .
3. Run ./cuda-tts.sh to generate speech

Server:

1. Download Pocket TTS models into the examples/docker/models/ folder
2. Run docker compose -f cuda-server.yml up
3. Run ./tts.sh to generate speech via a curl request

Known Issues

• The audio.cpp server does not terminate gracefully. The container cannot be stopped cleanly and must be killed (e.g. press Ctrl+C twice with Docker Compose). Proper signal handling is needed in the server.

Todo

Some open tasks before merging:

• Extend main documentation regarding Docker build and Docker examples
• Decide on naming convention for docker images (e.g. audiocpp vs audio.cpp)
• (Support Vulkan)