fix: close envs between tasks to prevent worker process accumulation

eval_policy_all never closed environments after each task completed,
causing AsyncVectorEnv worker processes to accumulate (N_tasks × n_envs).
This led to OOM, BrokenPipeError and EOFError on multi-task benchmarks.

Also fixes:
- AsyncVectorEnv compat in envs/utils.py (use get_attr/call instead of .envs)
- Tuple task handling in tokenizer_processor and lerobot_eval
- _LazyAsyncVectorEnv for deferred worker spawning in LIBERO

Made-with: Cursor

tests/envs/test_dispatch.py

@@ -90,7 +90,7 @@ def test_base_create_envs():
         envs = _Env().create_envs(n_envs=2)
         assert "_dispatch_base_test" in envs
         env = envs["_dispatch_base_test"][0]
-        assert isinstance(env, gym.vector.SyncVectorEnv)
+        assert isinstance(env, gym.vector.VectorEnv)
         assert env.num_envs == 2
         env.close()
     finally: