How to drop rollouts when the environment raises an error?

I have a pretty minimalistic setup of training PPO+PBT.

from gym import Env
from gym.spaces import Discrete
import ray
from ray.tune.schedulers import PopulationBasedTraining
from ray.tune import run
from ray.tune.registry import register_env


class CrashingEnv(Env):
    action_space = Discrete(2)
    observation_space = Discrete(3)

    def reset(self):
        return self.observation_space.sample()

    def step(self, action):
        raise RuntimeError()


def make_crashing_env(env_config) -> CrashingEnv:
    del env_config
    return CrashingEnv()


def test_train_crashing_env_ray_pbt_ppo():
    ray.init(local_mode=True)

    pbt = PopulationBasedTraining(metric="episode_reward_mean", mode="max")
    register_env("crashing_env", make_crashing_env)
    config = {
        "env": "crashing_env",
        "num_workers": 0,
    }
    run(
        run_or_experiment="PPO",
        name="crashing_env_ray_pbt_ppo",
        scheduler=pbt,
        config=config,
        )


if __name__ == "__main__":
    test_train_crashing_env_ray_pbt_ppo()

The environment deliberately raises an error on step(...). This would cause the rollout collection to fail and then the trial to fail.
How can I configure the training to drop/ignore the faulty rollout, remake the crashed environment in the rollout worker and continue training?