Fix Dockerfile for training environment

[Copilot]

The existing Dockerfile had multiple broken steps that would cause the build to fail and produce an incorrect training environment.

Fixes

• Broken npm install: COPY app*/package*.json + loop over app*/ matched nothing — replaced with explicit COPY service/package*.json service/ + COPY data/package*.json data/ and dedicated npm install per directory
• Conflicting JAX install: pip install -U "jax[cuda12]" ran outside the virtualenv and clobbered the jax-cuda13-pjrt/jax-cuda13-plugin pins already in requirements.txt — removed; requirements.txt is the sole source of truth for JAX+CUDA
• Wrong Makefile targets: make data → make datas, make protos/ → make protos
• Redundant proto step: standalone python protos/scripts/make_enums.py is already executed inside compile_protos.sh (invoked by make protos)
• Wrong CMD path: /start.sh → /app/start.sh

Improvements

• Replaced pip install virtualenv && virtualenv env with stdlib python3 -m venv env (adds python3-venv to apt layer)
• Consolidated protobuf-compiler into the single apt layer; added rm -rf /var/lib/apt/lists/*
• Protolint extracted to /tmp first, then only the binary moved to /usr/local/bin
• Combined two npm install -g calls into one layer
• Upgraded Node.js from 18.x → 20.x LTS
• Virtualenv explicitly activated in make datas and make protos RUN layers

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.