ci(docker): optimize Dockerfiles with multi-stage builds for faster build times

[pinglin]

Because

• The existing Dockerfiles had inefficient layer caching and sequential operations
• Build times were slow due to poor separation of dependencies and application code
• External downloads and package installations were not optimized for Docker's layer caching mechanism
• The production and development builds lacked proper multi-stage optimization

This commit

• Implements advanced multi-stage builds for both Dockerfile and Dockerfile.dev
• Separates build dependencies, binary dependencies, and runtime dependencies into distinct stages
• Optimizes Go build cache mounts to preserve build artifacts between builds
• Combines package installations into single layers to reduce image size
• Reorganizes external downloads (FFmpeg, ONNX Runtime, EasyOCR models) for better caching
• Creates dedicated stages for model artifacts and runtime setup
• Reduces final image size by excluding build-time dependencies from production images
• Improves incremental build performance through better layer separation and caching strategies