fix(rl): only set multiprocessing start method if not already set

mp.set_start_method() raises RuntimeError if called more than once
(e.g., when train_cli is invoked in a process that already set the
start method via pytest, Jupyter, or another entrypoint). Guard with
get_start_method(allow_none=True) check.

src/lerobot/rl/learner.py

@@ -125,7 +125,8 @@ def train_cli(cfg: TrainRLServerPipelineConfig):
     if not use_threads(cfg):
         import torch.multiprocessing as mp
 
-        mp.set_start_method("spawn")
+        if mp.get_start_method(allow_none=True) is None:
+            mp.set_start_method("spawn")
 
     # Use the job_name from the config
     train(