XTensorType.dims and .shape are silently mutable, leaving ndim out of sync

Reproduction

from pytensor.xtensor import xtensor

a = xtensor("a", dtype="float64", dims=("obs", "feature"))
b = xtensor("b", dtype="float64", dims=("feature",))

a.type.dims = ("g1", "g2", "obs", "feature")
a.type.shape = (None, None, None, None)

print(f"ndim={a.type.ndim}, len(dims)={len(a.type.dims)}")  # ndim=2, len=4

c = a @ b  # triggers shape rewrites; see traceback below

The @ operation triggers shape inference (shape.py:233) which iterates range(inp.type.ndim) to index into type.shape. Since ndim=2 but shape has 4 elements, indices 2 and 3 crash:

shape.py:233: tuple(input_shape_of(inp, j) for j in range(inp.type.ndim))
shape.py:275: if isinstance(var.type, HasShape) and var.type.shape[idx] is not None
IndexError: tuple index out of range

Expected behavior

dims and shape should be read-only properties. Mutating either should raise AttributeError.

Actual behavior

Both are plain instance attributes. ndim is computed once at __init__ and never updated.

Environment

pytensor 3.2.4 (PyPI) and 3.2.4 main branch — both affected

Start at the XTensorType definition and inspect shape.py at lines 233 and 275, where ndim and shape become inconsistent during shape inference. Reproduce the example, then verify that assigning to dims or shape raises AttributeError and that the matmul case no longer fails with an index error.