fix: Use backend/Dockerfile.backend instead of Dockerfile.codeengine

- Workflow was using Dockerfile.codeengine which has poetry install
- poetry install pulls CUDA PyTorch from poetry.lock (~6-8GB)
- backend/Dockerfile.backend has custom pip install for CPU-only PyTorch
- Also has email-validator fix for pydantic[email] syntax

This should resolve both CUDA libraries and email-validator issues.

Author: manavgup

.github/workflows/deploy_complete_app.yml

@@ -307,7 +307,7 @@ jobs:
         uses: docker/build-push-action@v6
         with:
           context: .
-          file: ./Dockerfile.codeengine
+          file: ./backend/Dockerfile.backend
           platforms: linux/amd64
           push: true
           # Versioning strategy: