Multi-core simulation: grain movement and heat across a worker pool (t88 phase 3)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

tools/perf.py

@@ -74,6 +74,9 @@ def report(result, baseline=None):
         print(f'{name:24}' + ''.join(f'{v:12.4f}' for v in values))
     for scene, d in result['scenes'].items():
         s = d['sim']
+        print(f"  {scene}: wall {s['total_ms']:.3f}, main occupied {s.get('main_thread_ms',s['total_ms']):.3f}, "
+              f"main compute {s.get('main_compute_ms',s['total_ms']):.3f}, polling {s.get('polling_ms',0):.3f} ms/tick; "
+              f"workers {s.get('workers',0)} (active peak {s.get('active_workers',0)})")
         print(f"  {scene}: awake {s['start']['awake']} -> {s['end']['awake']}, hot {s['start']['hot']} -> {s['end']['hot']} / {s['end']['chunks']}")
     print('\nReal loop: frame interval / CPU frame work / sim per frame, p50/p95/p99 ms')
     for scene, d in result['scenes'].items():
@@ -174,7 +177,7 @@ def main():
             raise RuntimeError('Refusing a software-rendered baseline')
         result['machine']['browser_version'] = browser.version
         def setup(scene):
-            return pg.evaluate('([s,w,h]) => perfScene(s,w,h)', [scene, *args.world])
+            return pg.evaluate('async ([s,w,h]) => { const state=perfScene(s,w,h); await startThreads(); return state; }', [scene, *args.world])
         for scene in scenes:
             print(f'Measuring {scene}...', file=sys.stderr, flush=True)
             setup(scene)