Accessing unobserved_value_vars changes model in place

[ricardoV94]

Description

Reported in https://discourse.pymc.io/t/missinginputerror-when-sampling-data/13426

import pytensor
import pymc as pm

with pm.Model() as m:
    theta1 = pm.Normal("theta1", 0, 1)
    theta2 = pm.Uniform("theta2", -theta1, theta1)

pytensor.dprint(theta2)
m.unobserved_value_vars
pytensor.dprint(theta2)  # Now includes the value var of theta1 in the graph!

theta2.eval()  # MissingInputError

Caused by the recent refactor in #6976
One of the replace functions that is now used causes an inplace change that was not caused by the previous one