fix: bump huggingface-hub to 0.20.3 to include split_torch_state_dict_into_shards

Dockerfile

@@ -37,7 +37,7 @@ RUN python -m pip install --upgrade pip && \
     # Speed up and lower RAM during model load (avoid upgrading hub)
     python -m pip install --no-deps accelerate && \
     # Ensure compatible hub version providing required symbols
-    python -m pip install --upgrade --force-reinstall "huggingface-hub==0.17.3" "diffusers==0.23.0"
+    python -m pip install --upgrade --force-reinstall "huggingface-hub==0.20.3" "diffusers==0.23.0"
 
 EXPOSE 7860