Dev fused bn functional

[MARD1NO]

增加 BatchNorm add Relu 的接口

一些未完成的细节：

1. cpu版本没实现，后续考虑补齐
2. affine = False参数不支持，需要在functional做较大改动，用户一般不会选择这样的选项

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 127.9ms (= 6394.0ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.1ms (= 7052.6ms / 50, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.10 (= 141.1ms / 127.9ms)

OneFlow resnet50 time: 74.3ms (= 3712.6ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.8ms (= 4189.7ms / 50, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.13 (= 83.8ms / 74.3ms)

OneFlow resnet50 time: 50.3ms (= 2513.7ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 61.4ms (= 3072.1ms / 50, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.22 (= 61.4ms / 50.3ms)

OneFlow resnet50 time: 44.9ms (= 2244.0ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.1ms (= 2457.2ms / 50, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.10 (= 49.1ms / 44.9ms)

OneFlow resnet50 time: 48.1ms (= 2403.8ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 43.3ms (= 2166.8ms / 50, input_shape=[1, 3, 224, 224])
❌ Relative speed: 0.90 (= 43.3ms / 48.1ms)

OneFlow resnet50 time: 152.1ms (= 7603.1ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 163.7ms (= 8186.2ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.08 (= 163.7ms / 152.1ms)

OneFlow resnet50 time: 100.8ms (= 5037.7ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 109.5ms (= 5474.5ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.09 (= 109.5ms / 100.8ms)

OneFlow resnet50 time: 84.0ms (= 4202.0ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 94.3ms (= 4713.4ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 94.3ms / 84.0ms)

OneFlow resnet50 time: 72.5ms (= 3624.2ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 77.2ms (= 3857.8ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.06 (= 77.2ms / 72.5ms)

OneFlow resnet50 time: 79.9ms (= 3993.9ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 62.6ms (= 3130.5ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 0.78 (= 62.6ms / 79.9ms)

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 128.0ms (= 6402.3ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.6ms (= 7078.1ms / 50, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 141.6ms / 128.0ms)

OneFlow resnet50 time: 74.3ms (= 3716.6ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.3ms (= 4166.0ms / 50, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.12 (= 83.3ms / 74.3ms)

OneFlow resnet50 time: 47.2ms (= 2361.2ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 61.0ms (= 3048.8ms / 50, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.29 (= 61.0ms / 47.2ms)

OneFlow resnet50 time: 40.9ms (= 2046.2ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 51.2ms (= 2560.0ms / 50, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.25 (= 51.2ms / 40.9ms)

OneFlow resnet50 time: 39.7ms (= 1987.2ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 43.3ms (= 2167.0ms / 50, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.09 (= 43.3ms / 39.7ms)

OneFlow resnet50 time: 152.5ms (= 7624.0ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 163.1ms (= 8154.7ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.07 (= 163.1ms / 152.5ms)

OneFlow resnet50 time: 100.1ms (= 5002.8ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 103.1ms (= 5153.2ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.03 (= 103.1ms / 100.1ms)

OneFlow resnet50 time: 79.8ms (= 3992.1ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 80.6ms (= 4032.1ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.01 (= 80.6ms / 79.8ms)

OneFlow resnet50 time: 69.6ms (= 3478.1ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 72.3ms (= 3616.5ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.04 (= 72.3ms / 69.6ms)

OneFlow resnet50 time: 71.5ms (= 3572.8ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 100.9ms (= 5043.4ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.41 (= 100.9ms / 71.5ms)