fix: split pip install layers, move precompute to runtime for Jaguar Space

The HF free-tier Docker builder has tight resource limits:
- Split torch/torchvision install into a separate layer
- Move dataset precomputation from build to runtime
- Increase healthcheck start-period to 600s for first-launch computation

Author: mnm-matin

spaces/jaguar-reid-megadescriptor-spherical/Dockerfile

@@ -25,21 +25,20 @@ ARG HYPERVIEW_VERSION=0.3.1
 
 # Install CPU-only PyTorch first (much smaller than default CUDA bundle),
 # then the released HyperView package with the ML extras.
-RUN pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu \
-    && pip install "hyperview[ml]==${HYPERVIEW_VERSION}" \
+RUN pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu
+RUN pip install "hyperview[ml]==${HYPERVIEW_VERSION}" \
     && python -c "import hyperview; print('hyperview', hyperview.__version__)"
 
 COPY --chown=user demo.py ./demo.py
 
 ENV HYPERVIEW_DATASETS_DIR=/home/user/app/demo_data/datasets \
     HYPERVIEW_MEDIA_DIR=/home/user/app/demo_data/media
 
-# Precompute at build time so the Space starts fast.
-RUN python -c "from demo import build_dataset; build_dataset()"
-
 EXPOSE 7860
 
-HEALTHCHECK --interval=30s --timeout=10s --start-period=120s --retries=3 \
+HEALTHCHECK --interval=30s --timeout=10s --start-period=600s --retries=3 \
     CMD curl -f http://localhost:7860/__hyperview__/health || exit 1
 
+# Dataset is built at first startup (model download + embedding computation).
+# After startup the data persists in demo_data/.
 CMD ["python", "demo.py"]