Add tests for scipy.sparse.csgraph and scipy.sparse.linalg

[mtsokol]

Hi @hameerabbasi,

Here's a PR that contains:

• Moved test_unary[sign((x_i is -0 or x_i == +0)) -> 0] to skips as this test is now flaky (it correctly failed in the PR but I see in the main branch it succeeded causing XPASS, so I guess --derandomize didn't get rid of all randomness)
• test_scipy_sparse_csgraph_dispatch with scipy.sparse.csgraph tests
• commented out a few linalg tests that require MAINT: sparse.linalg: adjust norm, eigs, and lsqr to pydata/sparse input scipy/scipy#20759 (and a new scipy release)
• test_scikit_learn_dispatch, skipped for now as it requires Add automatic conversion from pydata/sparse to scipy sparse counterparts scikit-learn/scikit-learn#29031

[github-actions]

Test Results

5 991 tests  +68   5 932 ✅ +40   9m 32s ⏱️ -11s
    1 suites ± 0      59 💤 +28 
    1 files   ± 0       0 ❌ ± 0 

Results for commit 1839415. ± Comparison against base commit 554cc7b.

This pull request removes 8 and adds 76 tests. Note that renamed tests count towards both.

sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Finch-coo-C]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Finch-coo-F]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Finch-csc-F]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Finch-csr-C]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Numba-coo-C]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Numba-coo-F]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Numba-csc-F]
sparse.tests.test_backends ‑ test_scipy_sparse_dispatch[BackendType.Numba-csr-C]

sparse.tests.test_backends ‑ test_scikit_learn_dispatch[BackendType.Finch-csc_matrix-csc-F]
sparse.tests.test_backends ‑ test_scikit_learn_dispatch[BackendType.Numba-csc_matrix-csc-F]
sparse.tests.test_backends ‑ test_scipy_breadth_first_tree[BackendType.Finch-coo_matrix-coo-F]
sparse.tests.test_backends ‑ test_scipy_breadth_first_tree[BackendType.Finch-csc_matrix-csc-F]
sparse.tests.test_backends ‑ test_scipy_breadth_first_tree[BackendType.Finch-csr_matrix-csr-C]
sparse.tests.test_backends ‑ test_scipy_breadth_first_tree[BackendType.Numba-coo_matrix-coo-F]
sparse.tests.test_backends ‑ test_scipy_breadth_first_tree[BackendType.Numba-csc_matrix-csc-F]
sparse.tests.test_backends ‑ test_scipy_breadth_first_tree[BackendType.Numba-csr_matrix-csr-C]
sparse.tests.test_backends ‑ test_scipy_connected_components[BackendType.Finch-coo_matrix-coo-F]
sparse.tests.test_backends ‑ test_scipy_connected_components[BackendType.Finch-csc_matrix-csc-F]
…

This pull request skips 3 and un-skips 1 tests.

sparse.numba_backend.tests.test_compressed ‑ test_reductions_float16[f4-None-sum-kwargs0]
sparse.numba_backend.tests.test_compressed ‑ test_reductions_float16[f8-None-sum-kwargs0]
sparse.numba_backend.tests.test_coo ‑ test_reductions_float16[f4-None-sum-kwargs0]

sparse.numba_backend.tests.test_coo ‑ test_reductions_float16[i4-axis4-mean-kwargs1]

♻️ This comment has been updated with latest results.

[hameerabbasi]

Thanks for the work here, @mtsokol! I've left some comments, mostly that the tests should be parametrized over function.

You can use a module-level PyTest fixture if you don't want to create the arrays over and over again.

[hameerabbasi]

Thanks for working on this @mtsokol!