Decompose nd sbp boxing

[lixinqi]

完成对称placement的nd sbp boxing。
大概实现思路：

1. 将nd sbp boxing降解为n个1d sbp boxing。
2. 依次执行这n个1d sbp boxing。

[lixinqi]

依赖分支 broadcast_consistent_shape和decorator_4_disable_recursive_boxing_call

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.5ms (= 6976.1ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 126.0ms (= 6300.1ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 139.5ms / 126.0ms)

PyTorch resnet50 time: 84.6ms (= 4231.2ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 72.8ms (= 3641.8ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.16 (= 84.6ms / 72.8ms)

PyTorch resnet50 time: 57.6ms (= 2878.3ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 46.2ms (= 2311.9ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.24 (= 57.6ms / 46.2ms)

PyTorch resnet50 time: 47.8ms (= 2391.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 40.2ms (= 2007.9ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.19 (= 47.8ms / 40.2ms)

PyTorch resnet50 time: 41.8ms (= 2089.2ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 40.5ms (= 2023.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.03 (= 41.8ms / 40.5ms)

[github-actions]

CI failed, removing label automerge

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 125.9ms (= 6294.3ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 136.5ms (= 6826.4ms / 50, input_shape=[16, 3, 224, 224])
Relative speed: 1.08 (= 136.5ms / 125.9ms)

OneFlow resnet50 time: 72.8ms (= 3638.0ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 82.1ms (= 4105.2ms / 50, input_shape=[8, 3, 224, 224])
Relative speed: 1.13 (= 82.1ms / 72.8ms)

OneFlow resnet50 time: 47.4ms (= 2371.3ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 60.9ms (= 3043.3ms / 50, input_shape=[4, 3, 224, 224])
Relative speed: 1.28 (= 60.9ms / 47.4ms)

OneFlow resnet50 time: 39.1ms (= 1955.4ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 48.1ms (= 2405.2ms / 50, input_shape=[2, 3, 224, 224])
Relative speed: 1.23 (= 48.1ms / 39.1ms)

OneFlow resnet50 time: 43.5ms (= 2174.2ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 43.5ms (= 2176.6ms / 50, input_shape=[1, 3, 224, 224])
Relative speed: 1.00 (= 43.5ms / 43.5ms)

OneFlow resnet50 time: 141.5ms (= 7075.0ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 159.0ms (= 7949.6ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
Relative speed: 1.12 (= 159.0ms / 141.5ms)

OneFlow resnet50 time: 91.8ms (= 4589.0ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 104.6ms (= 5229.6ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
Relative speed: 1.14 (= 104.6ms / 91.8ms)

OneFlow resnet50 time: 69.5ms (= 3473.3ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 75.2ms (= 3760.5ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
Relative speed: 1.08 (= 75.2ms / 69.5ms)

OneFlow resnet50 time: 67.1ms (= 3355.9ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.0ms (= 3250.9ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
Relative speed: 0.97 (= 65.0ms / 67.1ms)

OneFlow resnet50 time: 59.2ms (= 2959.3ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 63.4ms (= 3167.9ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
Relative speed: 1.07 (= 63.4ms / 59.2ms)