Fix test incompatibilities with numpy 2.5

[juanitorduz]

Description

CI on main has been failing since numpy 2.5 was picked up (e.g. this run on main) — 8 jobs fail, all reducing to three test-side incompatibilities:

1. tests/tensor/signal/test_conv.py::test_convolve2d (108 parametrizations, all tensor scipy+blas+pad jobs in every mode): the test seeded np.random.default_rng((172, kernel_shape, data_shape, sum(map(ord, mode)))) — a tuple containing tuples. numpy 2.5's SeedSequence rejects nested sequences with TypeError: SeedSequence does not accept nested sequences. Fixed by flattening the entropy tuple (still deterministic per parametrization).

2. tests/tensor/test_extra_ops.py::TestSearchsortedOp::test_searchsortedOp_on_right_side (NUMBA tensor *rest jobs) and

3. tests/link/numba/test_extra_ops.py::test_Searchsorted[a1-v1-left-None-None] (numba link job): both compared np.searchsorted against the pytensor/numba implementations on unsorted arrays. searchsorted on unsorted input is documented undefined behavior; numpy 2.5's reworked binary search now returns different (equally valid) results than numba's reimplementation, so the comparisons diverge. (CVM-mode jobs kept passing because the C implementation delegates to numpy's own routine.) Fixed by using sorted inputs — the same pattern test_searchsortedOp_on_sorted_input already uses.

Test-only changes; no library code touched. Verified locally with numpy 2.5.2: all 108 test_convolve2d params, the full TestSearchsortedOp suite, and all test_Searchsorted params pass (110 failed before).

These same failures currently show up on open PRs (e.g. #2375), where they are unrelated to the PR's changes.

Related Issue

• Closes #
• Related to failing CI on main (run 32712791280) and red checks on open PRs such as Fix numba failure on gradient of chained batched Convolve1d #2375

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):

[jessegrabowski]

i'd rather people not add the claude co-author line on their commits, but I am losing this battle.

[juanitorduz]

i'd rather people not add the claude co-author line on their commits, but I am losing this battle.

I can open a brand new PR (I did not know about this)

[jessegrabowski]

no need, just edit the commit message and force-push

[juanitorduz]

Done: removed the co-author line and force-pushed (here and on #2375).

[jessegrabowski]

thanks, feel free to merge when ci is green