Clean up loading code

- Centralized instantiation of the PEFT wrapper in `make_policy` for inference
  (e.g. in `lerobot-record`)
- Training a PEFT policy also sets `cfg.use_peft` so that all inference code loading
  the policy can rely on that attribute to identify if PEFT loading is needed
- Modified RTC example to also include PEFT policies. Mostly because this is an example
  I'm currently exploring.

examples/rtc/eval_with_real_robot.py

@@ -455,7 +455,14 @@ def demo_cli(cfg: RTCDemoConfig):
     if cfg.policy.type == "pi05" or cfg.policy.type == "pi0":
         config.compile_model = cfg.use_torch_compile
 
-    policy = policy_class.from_pretrained(cfg.policy.pretrained_path, config=config)
+    if config.use_peft:
+        from peft import PeftModel
+
+        policy = policy_class(config=config)
+        policy = PeftModel.from_pretrained(policy, cfg.policy.pretrained_path)
+
+    else:
+        policy = policy_class.from_pretrained(cfg.policy.pretrained_path, config=config)
 
     # Turn on RTC
     policy.config.rtc_config = cfg.rtc