feat(validation): chunk screenshots by max_images limit + fix websocket crash

- analyze_dashboard_multimodal now splits screenshots into chunks
  of max_images (from provider config) and sends them in parallel
- Results merged: worst status, deduped issues by (severity, msg, loc)
- New helper methods: _deduplicate_issues, _merge_chunk_results, _call_llm_for_images
- Plugin passes db_provider.max_images to the LLM client
- Report UI shows 'Chunked ×N' badge when analysis used multiple chunks
- i18n: added 'chunked' / 'По частям' key to validation.json
- Fix: isinstance(StopIteration) -> isinstance(_ws_exc, StopIteration)
  which crashed the websocket and broke task execution mid-flight
- Fix: update test mocks (_FakeLLMClient, _FakeScreenshotService)

backend/src/services/__tests__/test_llm_plugin_persistence.py

@@ -104,6 +104,10 @@ async def test_dashboard_validation_plugin_persists_task_and_environment_ids(
         async def capture_dashboard(self, _dashboard_id, _screenshot_path):
             return [], []
 
+        @staticmethod
+        def _cleanup_temp_files(_paths):
+            pass
+
     # endregion _FakeScreenshotService
 
     # region _FakeLLMClient [TYPE Class]
@@ -127,6 +131,22 @@ async def test_dashboard_validation_plugin_persists_task_and_environment_ids(
                 "issues": [],
             }
 
+        async def analyze_dashboard_multimodal(self, *_args, **_kwargs):
+            return {
+                "status": "PASS",
+                "summary": "Dashboard healthy",
+                "issues": [],
+            }
+
+        async def analyze_dashboard_text_batch(self, *_args, **_kwargs):
+            return {
+                "dashboards": [{
+                    "status": "PASS",
+                    "summary": "Dashboard healthy",
+                    "issues": [],
+                }],
+            }
+
     # endregion _FakeLLMClient
 
     # region _FakeNotificationService [TYPE Class]