Debug the Miniwob example

[NicolasAG]

• move the Miniwob agent example to domains/
• update miniwob config to not use tool calling. Instead it uses json step predictions
• scale up the number of env servers similarly to the number of actor vllm servers

[ollmer]

LGTM

[ollmer]

@rafapi I would like to merge this before my new actor PR, looks like it only affects miniwob rollouts, the only change affecting common actor logic is enforcing group size check:

                assert len(rollout_results) == attempts, (
                    f"Expected {attempts} rollouts, got {len(rollout_results)}"
                )

in actor.py. Are you ok to merge that?

[rafapi]

could we also merge main here?

[rafapi]

LGTM!