Dev Align torch avgpool

[MARD1NO]

对齐Pytorch的avgpool系列

和maxpool部分参数不同，因此这个PR暂时还未考虑将这两个pool相关代码共用

doctest

[BBuf]

这个限制为什么要存在呢

[MARD1NO]

pytorch有做对应的限制

[BBuf]

好的

[BBuf]

NDHW没听过，是不是写错了

[MARD1NO]

D -> depth。也有人写成NTCHW，T->time

[BBuf]

如果是支持NCHW和NHWC那么可以删掉这个注释，否则加assert

[MARD1NO]

这里暂时保留data_format参数，原因是AMP的dataformat会用到NHWC，如果有需要可以基于这个补充对应的NHWC版本kernel

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.6ms (= 6979.5ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 125.6ms (= 6278.7ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 139.6ms / 125.6ms)

PyTorch resnet50 time: 83.3ms (= 4165.4ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 72.8ms (= 3640.3ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 83.3ms / 72.8ms)

PyTorch resnet50 time: 57.3ms (= 2863.4ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 46.5ms (= 2326.3ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.23 (= 57.3ms / 46.5ms)

PyTorch resnet50 time: 49.2ms (= 2457.6ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 43.1ms (= 2152.8ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 49.2ms / 43.1ms)

PyTorch resnet50 time: 43.5ms (= 2172.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 38.6ms (= 1931.2ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 43.5ms / 38.6ms)