Also remove legacy `__div__`

We should instead install the GitHub App. (I don't have permissions to do this.)

See the announcement:
<https://about.readthedocs.com/blog/2025/06/announcing-our-github-app-beta/>

* Initial plan

* Remove InRange scalar op and add missing __rtruediv__ and __rfloordiv__

* Remove __rtruediv__ and __rfloordiv__ changes per review feedback

---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>

Using JAX Scan machinery to create MIT-SOT, SIT-SOT, and NIT-SOT buffers for us seems to be more performant than working directly on the pre-allocated buffers and reading/writing at every iteration. There is no machinery to work with MIT-MOT directly (just like in PyTensor user-facing Scan).

Co-authored-by: Jesse Grabowski <48652735+jessegrabowski@users.noreply.github.com>

Co-authored-by: Jesse Grabowski <48652735+jessegrabowski@users.noreply.github.com>

PyTensor Variables cannot be called `bool` upon

* Add mlx extra ops

* Add log_softmax dispatch for mlx

* Fix split bug, add tests

* Feedback

* Initial plan

* Rename ScalarMaximum/ScalarMinimum to Maximum/Minimum

* Apply ruff formatting fixes

* Remove custom names when class name matches desired name

* Remove scalar_ prefix from log1mexp, xlogx, xlogy0 and fix numba imports

* Fix xtensor test to use backward compat aliases in skip list

---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>

* Rename `mul` -> `multiply`

* Rename `sub` -> `subtract`

* Space... the final frontier

* `linalg.eig` always returns complex dtype

* Update Eig dispatch for Numba, Jax, and Pytorch backends

* Clean up `pytensor.linalg.expm` and related tests

* Add JAX dispatch for expm

* Implement L_op instead of grad in `Eigh`

---------
Co-authored-by: Jesse Grabowski <jesse.grabowski@readyx.com>

This isn't strictly needed but it's a more intuitive placement

Helpers before dispatchers

They are actually defined in tensor/math.py, but this is better than being in `basic.py`

It seems that libiomp5.lib has been renamed to libiomp5md.lib, so the detection was failing.
Add another search step with the new library name.

TODO: is there a more modern approach?
Co-authored-by: Luciano Paz <luciano.paz.neuro@gmail.com>

* Delete `sparse.sandbox`

* delete test_sp2.py

```bash
uvx gha-tools autoupdate ./.github/workflows/ --pin=all --version-strategy=specific --write
```