Make ConvLayerGrad usable on its own as a transposed convolution

[jessegrabowski]

A transposed convolution is the input gradient of a forward one, so ConvLayerGrad already computes it. Using it that way didn't work: it always computed the kernel gradient too, and differentiating it hit OpFromGraph's default pullback, which builds an anonymous op only numba can run.

It now returns either gradient alone, a rewrite drops whichever has no clients, and its own pullback is written as ConvLayer and one-sided ConvLayerGrad calls so every backend dispatches it. The ConvTranspose layers come next.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 89.28571% with 12 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.51%. Comparing base (e867086) to head (a8b3bca).

Files with missing lines	Patch %	Lines
pytensor_ml/dispatch/pytorch/conv.py	47.82%	12 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #102      +/-   ##
==========================================
- Coverage   97.89%   97.51%   -0.39%     
==========================================
  Files          59       59              
  Lines        2856     2938      +82     
==========================================
+ Hits         2796     2865      +69     
- Misses         60       73      +13     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.