When passing a size_t via binder, use 64-bits unconditionally

64-bits is almost always over-kill.  But it's easier and cleaner to change
the binder code to be accurate, than to rename all the the API parameter
types to be 32-bit.

Bug: 12381724
Change-Id: Ib8f198d814a2027760ef24e9e3feacee21a973b1
diff --git a/media/libmedia/IAudioFlingerClient.cpp b/media/libmedia/IAudioFlingerClient.cpp
index 3c0d4cf..1c299f7 100644
--- a/media/libmedia/IAudioFlingerClient.cpp
+++ b/media/libmedia/IAudioFlingerClient.cpp
@@ -55,7 +55,7 @@
             data.writeInt32(desc->samplingRate);
             data.writeInt32(desc->format);
             data.writeInt32(desc->channelMask);
-            data.writeInt32(desc->frameCount);
+            data.writeInt64(desc->frameCount);
             data.writeInt32(desc->latency);
         }
         remote()->transact(IO_CONFIG_CHANGED, data, &reply, IBinder::FLAG_ONEWAY);
@@ -85,7 +85,7 @@
                 desc.samplingRate = data.readInt32();
                 desc.format = (audio_format_t) data.readInt32();
                 desc.channelMask = (audio_channel_mask_t) data.readInt32();
-                desc.frameCount = data.readInt32();
+                desc.frameCount = data.readInt64();
                 desc.latency = data.readInt32();
                 param2 = &desc;
             }