Simplify matrix/tensor operations performed on scalars

[brandonwillard]

Ops like Solve and Det can be simplified/removed when their inputs are known to be scalars.

For example, the following graphs could be simplified but currently aren't:

import aesara
import aesara.tensor as at


A = at.tensor("floatX", name="A", shape=(1, 1))
b = at.tensor("floatX", name="b", shape=(1,))

z = at.slinalg.solve(A, b)
w = at.nlinalg.det(A)

z_w_fn = aesara.function([A, b], [z, w])

aesara.dprint(z_w_fn, print_type=True)
# Solve{assume_a='gen', lower=False, check_finite=True} [id A] <TensorType(float64, (1,))> 0
#  |A [id B] <TensorType(float64, (1, 1))>
#  |b [id C] <TensorType(float64, (1,))>
# Det [id D] <TensorType(float64, ())> 1
#  |A [id B] <TensorType(float64, (1, 1))>