fix: name the config key when step scheduler is missing step_size

Without step_size, StepLR raises 'missing 1 required positional argument',
which is Python-level phrasing that does not tell a config author the value
belongs under lr_scheduler_params. Raise with the config path instead.

Deliberately an error rather than a default: step_size determines the shape
of the entire lr curve, so a guessed value would train to completion and
quietly produce a worse result with no signal anything was wrong.

toolkit/scheduler.py

@@ -24,6 +24,11 @@ def get_lr_scheduler(
         # StepLR decays purely on step_size/gamma and has no notion of run
         # length, so drop the total_iters the trainer injects.
         kwargs.pop('total_iters', None)
+        if 'step_size' not in kwargs:
+            raise ValueError(
+                "lr_scheduler 'step' requires lr_scheduler_params.step_size "
+                "(number of steps between each lr decay)"
+            )
 
         return torch.optim.lr_scheduler.StepLR(
             optimizer, **kwargs