Have numpy use openblas for BLAS and LAPACK

[jjerphan]

No description provided.

[jjerphan]

I tried using threadpoolctl to test that OpenBLAS DSO are loaded at runtime, but I am getting errors.

 │ test_numpy.py .F....
 │ =================================== FAILURES ===================================
 │ _______________________ test_openblas_linked_at_runtime ________________________
 │     def test_openblas_linked_at_runtime():
 │         # Load NumPy's BLAS/LAPACK-backed extensions before introspecting.
 │         _ = np.dot(np.eye(2), np.ones(2))
 │     
 │         info = threadpool_info()
 │         openblas = [lib for lib in info if lib["internal_api"] == "openblas"]
 │ >       assert openblas, f"OpenBLAS not detected by threadpoolctl: {info}"
 │ E       AssertionError: OpenBLAS not detected by threadpoolctl: []
 │ E       assert []
 │ test_numpy.py:17: AssertionError
 │ =============================== warnings summary ===============================
 │ test_numpy.py::test_openblas_linked_at_runtime
 │   /lib/python3.13/site-packages/threadpoolctl.py:1129: UserWarning: Unable to import LDSO from pyodide_js._module. This should never happen.
 │     warnings.warn(

Given the UserWarning reported, I think threadpoolctl first needs to be adapted for this environment before it can report DSO and other pieces of information at runtime.