Add zero, constant, reflection and replication padding layers

[jessegrabowski]

Standalone padding layers, for the cases where the padding and the convolution want separate control. pt.pad already does the work, so these are mostly an argument surface over _pad_spatial — which moved out of the conv module on the way, since padding isn't convolution.

Two deliberate differences from torch. Padding is one entry per axis in input order, ((top, bottom), (left, right)), rather than torch's flat last-axis-first sequence; axis order is what the rest of the library already uses, and a flat 4-tuple gets an error saying so. And a reflection wider than the axis folds back and forth like numpy's, where torch refuses.

Part of #2.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 96.70330% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.47%. Comparing base (31666a4) to head (018e2fb).

Files with missing lines	Patch %	Lines
pytensor_ml/layers/padding.py	96.15%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #103      +/-   ##
==========================================
- Coverage   97.51%   97.47%   -0.04%     
==========================================
  Files          59       60       +1     
  Lines        2938     3011      +73     
==========================================
+ Hits         2865     2935      +70     
- Misses         73       76       +3     

☔ 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.