Sampling with generators as seeding no longer deterministic

[ricardoV94]

Description

This was revealed in pymc-labs/pymc-marketing#1269

import numpy as np
import pymc as pm

with pm.Model() as m:
    x = pm.Normal("x")

    post1 = pm.sample(tune=10, draws=10, random_seed=np.random.default_rng(42)).posterior
    post2 = pm.sample(tune=10, draws=10, random_seed=np.random.default_rng(42)).posterior
assert post1.equals(post2), (post1["x"].mean().item(), post2["x"].mean().item())
# AssertionError: (0.22006495904628473, -0.31090965213192406)

[ricardoV94]

@lucianopaz seems to be caused by #7508

[lucianopaz]

This might be fixed by the second, third, and fourth commits from #7540. This might be caused by that bug from numpy<2.0 where the random generator was not copied properly.