fix(ci): add HF_TOKEN and model caching, remove uv.lock

- Add HF_TOKEN environment variable to CI workflow for authenticated HuggingFace downloads
- Add HuggingFace model caching to avoid repeated downloads and HTTP 429 errors
- Set TRANSFORMERS_CACHE and HF_HOME environment variables for consistent model paths
- Remove uv.lock that was accidentally committed in rollback commit
- Add uv.lock to .gitignore to prevent future commits

This fixes the CI failure caused by HuggingFace rate limiting when multiple tests load the sentence-transformers model.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: HumanBean17

.github/workflows/test.yml

@@ -51,10 +51,22 @@ jobs:
       - name: Check generated edge navigation doc
         if: steps.changes.outputs.code == 'true'
         run: python scripts/generate_edge_navigation.py --check
+      - name: Cache HuggingFace models
+        if: steps.changes.outputs.code == 'true'
+        uses: actions/cache@v4
+        with:
+          path: ~/.cache/huggingface
+          key: hf-${{ hashFiles('**/pyproject.toml') }}-v1
+          restore-keys: |
+            hf-${{ hashFiles('**/pyproject.toml') }}-
+            hf-
       - name: Run tests
         if: steps.changes.outputs.code == 'true'
         env:
           JAVA_CODEBASE_RAG_RUN_HEAVY: "0"
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+          TRANSFORMERS_CACHE: ~/.cache/huggingface
+          HF_HOME: ~/.cache/huggingface
         run: pytest tests -v
       - name: Skip tests (docs-only)
         if: steps.changes.outputs.code != 'true'

.gitignore

@@ -56,3 +56,6 @@ cocoindex_java_lance.db/
 .env
 .env.*
 tmp/
+
+# uv lock file (should not be committed)
+uv.lock