feat(perception): add frozen yolox provider

tests/test_perception_graph.py

@@ -50,6 +50,7 @@ from k1link.perception.providers import (
     SourcePacket,
 )
 from k1link.perception.recorded_source import (
+    DecodedRecordedSource,
     RecordedRavnoves00Source,
     RecordedSourceError,
     ReplayPacing,
@@ -529,6 +530,29 @@ def test_recorded_source_rejects_timeline_mismatch(tmp_path: Path) -> None:
         list(source.packets(Event()))
 
 
+def test_decoded_recorded_source_attaches_images_without_detector_logic(tmp_path: Path) -> None:
+    camera_path, timeline_path = _write_recorded_fixture(tmp_path)
+    source = RecordedRavnoves00Source(
+        camera_index_path=camera_path,
+        source_pack_path=timeline_path,
+        expected_frame_count=2,
+        expected_source_pack_sha256=None,
+    )
+
+    class Decoder:
+        def frames(self, stop_event: Event) -> Iterator[np.ndarray]:
+            for value in (3, 7):
+                if stop_event.is_set():
+                    return
+                yield np.full((600, 800, 3), value, dtype=np.uint8)
+
+    packets = list(DecodedRecordedSource(source=source, decoder=Decoder()).packets(Event()))
+    assert len(packets) == 2
+    assert isinstance(packets[0].image_payload, np.ndarray)
+    assert int(packets[0].image_payload[0, 0, 0]) == 3
+    assert int(packets[1].image_payload[0, 0, 0]) == 7
+
+
 def test_camera_only_path_never_invents_metric_occupancy_or_free_space() -> None:
     result = _graph(_Source((_packet(0, lidar=False),))).run()
     delivery = result.deliveries[0]