Fix test_create_table_numpy to run without pandas in the numpy: 1 CI leg

[aayushgupta7725]

Fixes #872

The numpy: 1 leg of the test matrix in .github/workflows/test.yml
installs numpy but deliberately not pandas, to check numpy-only
scenarios. However, the only test meant to run there,
test_create_table_numpy, was decorated with
@pytest.mark.skipif(pd is None, ...) and built its test data via
pd.DataFrame(...).to_dict(orient="records") — so it required pandas
and was silently skipped in every numpy: 1 job, every time, since
2020.

This PR rewrites the test to build its rows directly with numpy's own
constructors (np.int8, np.uint64, np.float16, etc.), using
.item() to convert each value to a plain Python type the same way
pandas' to_dict() already does internally. This removes the pandas
dependency from the test while preserving exactly what it verifies —
that sqlite-utils can build a table and insert data originating from
various numpy dtypes.

Changes:

• Rewrote test_create_table_numpy in tests/test_create.py to use
  numpy directly instead of pd.DataFrame
• Changed the skipif condition from pd is None to np is None
• Removed the now-unused import pandas as pd block from
  tests/test_create.py, since nothing else in the file references
  pd

Verified locally: with pandas uninstalled and numpy installed
(matching the numpy: 1 CI leg), the test now runs and passes rather
than being skipped.

---

📚 Documentation preview 📚: https://sqlite-utils--875.org.readthedocs.build/en/875/