XTensorVariable indexing update

[ricardoV94]

This builds on top of #1429 to support index assignment and increment x[idx] = y and x[idx] += y, following xarray indexing and broadcasting semantics.

As with regular TensorVariables, due to immutability/hashability constraints of PyTensor, it is not possible to use the python native operations, so one has to write z = x[idx].set(y) and z = x[idx].inc(y)

---

📚 Documentation preview 📚: https://pytensor--1438.org.readthedocs.build/en/1438/

[OriolAbril]

I was also unable to find a combination of indexes/setter data that I think should work but doesn't.

This one already works but I'd suggest adding as a test case too:

x = xtensor("x", shape=(5, 5), dims=("a", "b"))
y = xtensor("y", shape=(3,), dims=("d",))
idx = xtensor("idx", dtype=int, shape=(None,), dims=("d",))

# define "d" dimension by slicing the "a" dimension
# so we can set y into x
orthogonal_update1 = x[idx].set(y)
fn = xr_function([x, idx, y], [orthogonal_update1])

x_test = np.abs(xr_random_like(x))
y_test = -np.abs(xr_random_like(y))
idx_test = DataArray([0, 2, 3], dims=("d",))

result = fn(x_test, idx_test, y_test)
x_test[idx_test] = y_test
expected_result = [x_test]
xr_assert_allclose(result, expected_result)

[ricardoV94]

This one already works but I'd suggest adding as a test case too:

Will do!