Description

The QRFull Op in pytensor.tensor.nlinalg does not implement infer_shape, which means it cannot be vectorized with blockwise when mode=NUMBA. It should be easy to implement -- just have to check the mode property and tailor the shapes accordingly.