Fix CI timeout: cache HuggingFace tokenizer and increase test timeout

Tests download the GPT-2 tokenizer from HuggingFace on first run.
On CI runners this can stall and exceed the 120s timeout. Added
HuggingFace cache step and increased timeout to 300s.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: brian-c-moore

.github/workflows/tests.yml

@@ -26,6 +26,12 @@ jobs:
           python -m pip install --upgrade pip
           pip install -e ".[dev]"
 
+      - name: Cache HuggingFace models
+        uses: actions/cache@v4
+        with:
+          path: ~/.cache/huggingface
+          key: hf-${{ runner.os }}-${{ matrix.python-version }}
+
       - name: Run unit tests
         run: |
-          python -m pytest tests/ -v --timeout=120 -k "not test_integration and not test_trained_model"
+          python -m pytest tests/ -v --timeout=300 -k "not test_integration and not test_trained_model"