`ContinuousCardinalityConstraint` not considered in hybrid spaces

[AdrianSosic]

Noticed by @myrazma

The problem is that, in hybrid optimization, we directly call BoTorch's optimize_acqf_mixed routine, bypassing our cardinality constraint machinery.

Here a minimal reproducing example:

from baybe import Campaign
from baybe.constraints import ContinuousCardinalityConstraint
from baybe.parameters import CategoricalParameter, NumericalContinuousParameter
from baybe.searchspace import SearchSpace
from baybe.targets.numerical import NumericalTarget
from baybe.utils.dataframe import add_fake_measurements

parameters = [
    NumericalContinuousParameter("p1", (0, 1)),
    NumericalContinuousParameter("p2", (0, 1)),
    CategoricalParameter("p_cat", ["A", "B", "C"]),
]
constraints = [ContinuousCardinalityConstraint(["p1", "p2"], max_cardinality=1)]
searchspace = SearchSpace.from_product(parameters, constraints)
objective = NumericalTarget("t", "MAX").to_objective()
campaign = Campaign(searchspace, objective)

recommendations = campaign.recommend(10)
add_fake_measurements(recommendations, campaign.targets)
campaign.add_measurements(recommendations)

recommendations = campaign.recommend(10)
print(recommendations)

Output looks like:

      p_cat        p1        p2
index                          
1         B  0.046313  0.009410
1         B  0.010788  0.283344
1         B  0.030576  0.076704
1         B  0.012316  0.405818
1         B  0.857327  0.101991
1         B  0.821122  0.000000
1         B  0.870719  0.000000
1         B  0.000000  0.251797
1         B  0.028825  0.327676
1         B  0.900114  0.110927