Implement gradient for vector repetitions

[ricardoV94]

Nobody asked for it, and nobody probably ever needed it (it exists since 2012), but it was a fun challenge.

Also cleaned up a bit the implementation by not allowing negative axis on the Op itself like we do in other cases (the helper repeat handles the conversion to positive axis)

doctest fails due to an unrelated print that should be solved by #1193

---

📚 Documentation preview 📚: https://pytensor--1192.org.readthedocs.build/en/1192/

[codecov]

Codecov Report

Attention: Patch coverage is 92.30769% with 3 lines in your changes missing coverage. Please review.

Project coverage is 82.26%. Comparing base (4fa9bb8) to head (2543387).
Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
pytensor/tensor/extra_ops.py	92.10%	1 Missing and 2 partials ⚠️

❌ Your patch status has failed because the patch coverage (92.30%) is below the target coverage (100.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1192      +/-   ##
==========================================
- Coverage   82.27%   82.26%   -0.02%     
==========================================
  Files         186      186              
  Lines       48066    48069       +3     
  Branches     8633     8633              
==========================================
- Hits        39546    39543       -3     
- Misses       6360     6366       +6     
  Partials     2160     2160              

Files with missing lines	Coverage Δ
pytensor/link/vm.py	92.00% <100.00%> (ø)
pytensor/tensor/extra_ops.py	88.06% <92.10%> (+0.05%)	⬆️

... and 1 file with indirect coverage changes

[jessegrabowski]

Needs some small cleanup, otherwise cool!

[jessegrabowski]

What if I'm working in pure pytensor? I have to import numpy just to read our docs?

[ricardoV94]

you go to the docs :)

[ricardoV94]

We could add links like JAX does, I think we already have cross-ref so it should work the same way: https://github.com/jax-ml/jax/blob/e7acb20ea2ba037a0e82134a6b1bcf35430d28ca/jax/_src/numpy/lax_numpy.py#L6980

Then the docs have a clickable link: https://docs.jax.dev/en/latest/_autosummary/jax.numpy.repeat.html

[jessegrabowski]

Suggested change

	a: tensor_like
	a: TensorLike

I think the pycharm linter checks these for valid types. There's no type called tensor_like, and it's also not fully human readable; kind of a weird middle ground

[ricardoV94]

I thought we had a glossary like in PyMC: https://www.pymc.io/projects/docs/en/stable/glossary.html#term-tensor_like

In which case we should use tensor_like in the docs

[ricardoV94]

CC @OriolAbril