Fix inconsistent t_step_start access in run_restart Phase 1

Phase 1 used orig.get('t_step_start', 0) to compute t_step_save, while
Phase 2 and the mid_step guard both use orig['t_step_start'] (hard key
access). t_step_start is always present (provided by BASE_CFG), so the
.get() default silently produces a wrong t_step_save if the key is ever
absent. Use consistent direct access throughout.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: sbryngelson

toolchain/mfc/test/case.py

@@ -159,7 +159,7 @@ def run_restart(self, targets, gpus):
 
             # Phase 1: Run to midpoint (generates restart data)
             self.params = {**orig, 't_step_stop': mid_step,
-                           't_step_save': mid_step - orig.get('t_step_start', 0)}
+                           't_step_save': mid_step - orig['t_step_start']}
             self.create_directory()
             result1 = self.run(targets, gpus)
             if result1.returncode != 0: