Fix Blockwise and RandomVariable in Numba with repeated arguments

[ricardoV94]

To implement Blockwise and RandomVariables in numba we use an OpWithCoreShape OpFromGraph subclass, that extends the original node with a core_shape input. This approach would fail whenever the original node had repeated inputs, as OpFromGraph raises by default when that's the case.

For now I think simply allowing OpFromGraph to ignore those disconnected inputs should work fine. We can revisit later and use a node.clone_with_new_inputs or whatever is called to mask the fact the inputs are identical. I have a 90% confidence this won't be needed so I'm trying the simple approach until reality proves it wrong.

---

📚 Documentation preview 📚: https://pytensor--1222.org.readthedocs.build/en/1222/

[jessegrabowski]

I don't get it, where's the repeated arg?

[ricardoV94]

matmul with x twice

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.03%. Comparing base (e25e8a2) to head (2e036b2).
Report is 4 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1222   +/-   ##
=======================================
  Coverage   82.03%   82.03%           
=======================================
  Files         188      188           
  Lines       48535    48537    +2     
  Branches     8676     8676           
=======================================
+ Hits        39814    39816    +2     
  Misses       6558     6558           
  Partials     2163     2163           

Files with missing lines	Coverage Δ
pytensor/tensor/blockwise.py	85.64% <100.00%> (+0.13%)	⬆️