Create canonicalization for sub(x, neg(y))

[ricardoV94]

Seems like we are missing a canonicalization for sub(x, neg(y)) -> add(x, y) :

import aesara
import aesara.tensor as at

x = at.scalar("x")
y = at.scalar("y")
z = x - (-y)

f = aesara.function([x, y], z)
aesara.dprint(f)

Elemwise{Composite{(i0 - (-i1))}} [id A] ''   0
 |x [id B]
 |y [id C]

It does work with constants

z = 5 - (-x)

f = aesara.function([x], z)
aesara.dprint(f)

Elemwise{add,no_inplace} [id A] ''   0
 |TensorConstant{5.0} [id B]
 |x [id C]

And we have one for double negation:

z = - (-x)

f = aesara.function([x], z)
aesara.dprint(f)

DeepCopyOp [id A] 'x'   0
 |x [id B]

Originally posted by @ricardoV94 in #473 (comment)

[ricardoV94]

We might also want to canonicalize -x + y -> y - x.

Not sure if it is a faster operation, but some lazy benchmarks on Colab suggest so. In any case it might be worth it just as a canonicalization.