Optimize Docker image sizes for FL client and server containers

[abdulmeLINK]

The current Docker images for FL client and server components are excessively large due to several optimization opportunities:

Current Issues

1. Large ML Dependencies: Both images include PyTorch (torch==1.13.1) and TorchVision (torchvision==0.14.1), which significantly increase image size
2. Unnecessary Files: The entire src/ directory is copied including pycache and potentially unused files
3. No Multi-stage Builds: Single-stage builds include build dependencies in final images
4. Excessive Networking Tools: Multiple networking utilities installed that may not all be needed
5. Cache Not Cleaned: pip and apt caches not properly cleaned

Impact

• Increased storage requirements
• Slower image pulls and deployments
• Higher bandwidth usage in GNS3 environments
• Longer build times

Proposed Solutions

1. Multi-stage Builds: Use separate build and runtime stages
2. Selective Dependencies: Only install required ML libraries or use lighter alternatives
3. Minimal Base Images: Consider alpine-based images or distroless
4. Clean Copy Operations: Use .dockerignore and exclude unnecessary files
5. Runtime Optimizations: Use python -m py_compile for bytecode, remove pip cache

Files to Update

• docker/flopynet_fl_client.Dockerfile
• docker/flopynet_fl_server.Dockerfile
• docker/requirements/fl-client-requirements.txt
• docker/requirements/fl-server-requirements.txt

Acceptance Criteria

• Reduce image size by at least 50%
• Maintain all required functionality
• No breaking changes to container interfaces
• Document size improvements in build logs