make flow._C.local_all_reduce sync lanuched

[daquexian]

暂时把 flow._C.local_all_reduce 设置成 sync launched，这样子不会出现两卡结果不一致的问题，等待 allocator per stream 功能实现之后恢复

[github-actions]

CI failed, removing label automerge

[github-actions]

CI failed, removing label automerge

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 127.8ms (= 6388.8ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.1ms (= 7104.8ms / 50, input_shape=[16, 3, 224, 224])
Relative speed: 1.11 (= 142.1ms / 127.8ms)

OneFlow resnet50 time: 74.3ms (= 3712.8ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 81.1ms (= 4053.5ms / 50, input_shape=[8, 3, 224, 224])
Relative speed: 1.09 (= 81.1ms / 74.3ms)

OneFlow resnet50 time: 47.9ms (= 2395.9ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 60.0ms (= 3001.3ms / 50, input_shape=[4, 3, 224, 224])
Relative speed: 1.25 (= 60.0ms / 47.9ms)

OneFlow resnet50 time: 43.5ms (= 2176.8ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 47.7ms (= 2383.5ms / 50, input_shape=[2, 3, 224, 224])
Relative speed: 1.09 (= 47.7ms / 43.5ms)

OneFlow resnet50 time: 42.9ms (= 2142.5ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.6ms (= 2027.8ms / 50, input_shape=[1, 3, 224, 224])
Relative speed: 0.95 (= 40.6ms / 42.9ms)