checkpoint: error if --checkpoint.load_step is specified but not found (#1311)

Currently when `--checkpoint.load_step=xxx` is specified (not -1), and
no corresponding `checkpoint/step-xxx` folder is found, it will
~silently~ start from random initialization. User have to check the log
for a single line `Loading checkpoint step ....` to make sure it's
working.
I think that's not very intuitive and maybe explicitly error-out would
be better.

Author: lkhphuc

torchtitan/components/checkpoint.py

@@ -445,7 +445,9 @@ def load(self, step: int = -1) -> bool:
             checkpoint_id = self._create_checkpoint_id(step)
 
             if not os.path.isdir(checkpoint_id):
-                return False
+                raise FileNotFoundError(
+                    f"--checkpoint.load_step={step} but checkpoint {checkpoint_id} is not found."
+                )
 
         logger.info(f"Loading the checkpoint from {checkpoint_id}.")
         begin = time.monotonic()