[Bug]: [RunInference] max_models_per_worker_hint is not enforced #40468

from apache_beam.internal import pickler
from apache_beam.ml.inference import base
class Model:
  def predict(self, x):
    return x
class Handler(base.ModelHandler):
  def load_model(self):
    return Model()
  def run_inference(self, batch, model, inference_args=None):
    return [model.predict(x) for x in batch]
mhs = [base.KeyModelMapping([k], Handler()) for k in ('a', 'b', 'c')]
keyed_handler = base.KeyedModelHandler(mhs, max_models_per_worker_hint=1)
# RunInference shares one _ModelHandlerManager per transform across all
# DoFn instances (and processes) via MultiProcessShared.
manager = keyed_handler.load_model()
# 5 DoFn instances in one process (harness threads, re-created bundle
# processors); each deserializes its own copy of the model handler.
for _ in range(5):
  handler_copy = pickler.roundtrip(keyed_handler)
  handler_copy.override_metrics('ns')
  handler_copy.run_inference([('a', 1), ('b', 2), ('c', 3)], manager)
print('model limit:', manager._max_models)  # 5, expected 1
print('models in memory:', len(manager._tag_map))  # 3