Right now the final command buffer in swapchain_impl_vk.cc is skipped because it is untracked. We should find a way to include that, as we might be skipping a decent amount of work in these measurements. See how GPU frame time dropped in https://flutter-flutter-perf.skia.org/e/?begin=1697570839&end=1697819332&numCommits=20&queries=device_type%3DPixel_7_Pro%26sub_result%3D90th_percentile_frame_rasterizer_time_millis%26sub_result%3D90th_percentile_gpu_frame_time%26sub_result%3Daverage_frame_rasterizer_time_millis%26sub_result%3Daverage_gpu_frame_time%26sub_result%3Dworst_frame_rasterizer_time_millis%26test%3Danimated_advanced_blend_perf__timeline_summary&requestType=0&selected=commit%3D37507%26name%3D%252Carch%253Dintel%252Cbranch%253Dmaster%252Cconfig%253Ddefault%252Cdevice_type%253DPixel_7_Pro%252Cdevice_version%253Dnone%252Chost_type%253Dlinux%252Csub_result%253Daverage_frame_rasterizer_time_millis%252Ctest%253Danimated_advanced_blend_perf__timeline_summary%252C

after the methodology switched.