Unwrap picture ids in the RtpFrameReferencerFinder.

First CL to avoid working with wrapping picture ids. After the references of
a frame has been determined they are then unwrapped.

BUG=webrtc:7874

Review-Url: https://codereview.webrtc.org/2985283002
Cr-Commit-Position: refs/heads/master@{#19663}
diff --git a/webrtc/modules/video_coding/frame_object.h b/webrtc/modules/video_coding/frame_object.h
index 5299ed8..90c1c00 100644
--- a/webrtc/modules/video_coding/frame_object.h
+++ b/webrtc/modules/video_coding/frame_object.h
@@ -49,14 +49,14 @@
   // The tuple (|picture_id|, |spatial_layer|) uniquely identifies a frame
   // object. For codec types that don't necessarily have picture ids they
   // have to be constructed from the header data relevant to that codec.
-  uint16_t picture_id;
+  int64_t picture_id;
   uint8_t spatial_layer;
   uint32_t timestamp;
 
   // TODO(philipel): Add simple modify/access functions to prevent adding too
   // many |references|.
   size_t num_references;
-  uint16_t references[kMaxFrameReferences];
+  int64_t references[kMaxFrameReferences];
   bool inter_layer_predicted;
 };