Enable UBSan float-cast-overflow warnings and fix existing ones

webrtc/modules/audio_processing/level_controller/level_controller.cc

Index: webrtc/modules/audio_processing/level_controller/level_controller.cc
diff --git a/webrtc/modules/audio_processing/level_controller/level_controller.cc b/webrtc/modules/audio_processing/level_controller/level_controller.cc
index 028501c5c03bd3e3904710d275f65855d02be6f1..6edae5c38a8039b72c3df2a62a1815f1bda0ed7a 100644
--- a/webrtc/modules/audio_processing/level_controller/level_controller.cc
+++ b/webrtc/modules/audio_processing/level_controller/level_controller.cc
@@ -49,7 +49,7 @@ float FrameEnergy(const AudioBuffer& audio) {
   for (size_t k = 0; k < audio.num_channels(); ++k) {
     float channel_energy =
         std::accumulate(audio.channels_const_f()[k],
-                        audio.channels_const_f()[k] + audio.num_frames(), 0,
+                        audio.channels_const_f()[k] + audio.num_frames(), 0.f,

[saza WebRTC, 2017/09/05 09:35:20]

Good find. Fyi, aleloi@ or hlundin@ are the least busy OWNERs of
audio_processing/ who can lgtm this.

                         [](float a, float b) -> float { return a + b * b; });
     energy = std::max(channel_energy, energy);
   }