M10: ProcessingRunner skeleton

- Add modules/runtime/runner.py with ProcessingRunner and ProcessingRequest
- Wire process_images to delegate through runner (internal only)
- Add test/quality/test_processing_runner.py contract test

Behavior-preserving. Zero blast radius. All callers unchanged.

Made-with: Cursor

modules/processing.py

@@ -844,7 +844,10 @@ def process_images(p: StableDiffusionProcessing) -> Processed:
             sd_samplers.fix_p_invalid_sampler_and_scheduler(p)
 
             with profiling.Profiler():
-                res = process_images_inner(p)
+                from modules.runtime.runner import ProcessingRunner, ProcessingRequest
+                runner = ProcessingRunner()
+                request = ProcessingRequest(p)
+                res = runner.run(request)
     finally:
         sd_models.apply_token_merging(p.sd_model, 0)