Remove example usage of deprecating trlx.train dataset arg

[jon-tow]

Quick fix to remove all dataset args to trlx.train to avoid promoting deprecating usage.

• If dataset is provided, the default behavior is to load the SFT config even though ILQL could be the intended method (see simulacra notebook).
  

  trlx/trlx/trlx.py

  Lines 57 to 62 in 165422d

  	if reward_fn:
  	config = TRLConfig.load_yaml("configs/ppo_config.yml")
  	elif rewards:
  	config = TRLConfig.load_yaml("configs/ilql_config.yml")
  	else:
  	config = TRLConfig.load_yaml("configs/sft_config.yml")

[cat-state]

Good catch!