Numba obj fallback fails for Blockwise with scalar outputs

[ricardoV94]

Description

import numpy as np
import pytensor
import pytensor.tensor as pt
from pytensor.graph.basic import Apply
from pytensor.scalar import ScalarOp
from pytensor.tensor.elemwise import Elemwise
from pytensor.tensor.blockwise import Blockwise


class ScalarOp(ScalarOp):
    def make_node(self, x):
        return Apply(self, [x], [x.type()])

    def perform(self, node, inputs, outputs):
        [x] = inputs
        outputs[0][0] = x + 1

x = pt.vector("x")
y = Elemwise(ScalarOp())(x)
fn = pytensor.function([x], y, mode="NUMBA")
# UserWarning: Numba will use object mode to run ScalarOp's perform method
np.testing.assert_allclose(fn(np.zeros((3,))), [1, 1, 1])

x = pt.vector("x")
z = Blockwise(ScalarOp(), signature="()->()")(x)
# UserWarning: Numba will use object mode to run ScalarOp's perform method
fn = pytensor.function([x], z, mode="NUMBA")
fn(np.zeros((3,)))  # TypeError: can't unbox array from PyObject into native value.  The object maybe of a different type

Blockwise is different than Elemwise in that it gets wrapped in a BlockwiseWithCoreShape, which is an OpFromGraph. So the failure may be more general to OpFromGraph