Incorrect shape inferrence for blockwise dot

[aseyboldt]

Description

It seems some part of the shape inference for blockwise gives incorrect results:

import pytensor
import pytensor.tensor as pt

x = pt.tensor("x", shape=(19, 19))
q = np.random.randn(20, 19)
(q @ x).type
# TensorType(float64, shape=(19, 19))

This gives me (19, 19) on main, when it should be (20, 19).

If I wrap q into a tensor variable I get the correct result:

import pytensor
import pytensor.tensor as pt

x = pt.tensor("x", shape=(19, 19))
q = pt.as_tensor_variable(np.random.randn(20, 19))
(q @ x).type
# TensorType(float64, shape=(20, 19))

[ricardoV94]

Should be solved by #465