Limit inter-layer prediction to key pictures.

This unconditionally limits usage of VP9 SVC inter-layer prediction to
frames of key picture.

Ideally we would like to let an application to configure SVC options.
But currently there is no API for this.

Bug: none
Change-Id: I21a84dafc946be122514d5b6bf327b65251f1115
Reviewed-on: https://webrtc-review.googlesource.com/76640
Commit-Queue: Sergey Silkin <ssilkin@webrtc.org>
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#23224}

media/engine/webrtcvideoengine.cc

@@ -479,6 +479,9 @@ WebRtcVideoChannel::WebRtcVideoSendStream::ConfigureVideoEncoderSettings(
           num_spatial_layers, kConferenceMaxNumSpatialLayers);
       vp9_settings.numberOfTemporalLayers = std::min<unsigned char>(
           num_temporal_layers, kConferenceMaxNumTemporalLayers);
+
+      // Limit inter-layer prediction to key pictures.
+      vp9_settings.interLayerPred = webrtc::InterLayerPredMode::kOnKeyPic;
     }
 
     // VP9 denoising is disabled by default.