`aet.switch` fails `local_useless_switch` optimizations when used with certain conditions.

[kc611]

MWE:

import aesara
import aesara.tensor as aet
import numpy as np

aesara.config.floatX = 'float32'

cond = aet.as_tensor_variable(np.array([True]))
ift = aet.as_tensor_variable(np.array([0]))
iff = aet.as_tensor_variable(np.array([0]))

a = aet.switch(cond, ift, iff)
b = a.eval() # Throws Error while optimizations

The solutions also looks straightforward, the issue is probably that when cond=True, it fails to go through this particular branch:

aesara/aesara/tensor/basic_opt.py

Lines 2538 to 2540 in 29032f3

	if (isinstance(cond, np.ndarray) and cond.ndim == 0) or isinstance(
	cond, np.number
	):

Adding an additional condition like or cond in [True, False] should fix this issue.

Note that this only happens when floatX is set to float32.