add formula of CombinedMarginLoss

[marigoold]

此PR完成了：
给CombinedMarginLoss增加了公式，来表明每个参数的作用，并且增加了ArcFace、CosFace和A-Softmax的note

[doombeaker]

@kevinJhz 麻烦江同学 review 这个PR。中间有疑问可以和作者 @marigoold 交流请教。

需要先弄懂这个算子的功能，可能前置条件本身就一定挑战（比如可能需要去看下 kernel 实现）

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8206/

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 129.3ms (= 12926.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 146.3ms (= 14628.4ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.13 (= 146.3ms / 129.3ms)

OneFlow resnet50 time: 75.5ms (= 7550.4ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.3ms (= 8433.3ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.12 (= 84.3ms / 75.5ms)

OneFlow resnet50 time: 48.7ms (= 9742.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 61.1ms (= 12210.9ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.25 (= 61.1ms / 48.7ms)

OneFlow resnet50 time: 39.5ms (= 7897.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 42.6ms (= 8518.0ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.08 (= 42.6ms / 39.5ms)

OneFlow resnet50 time: 34.4ms (= 6871.2ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 35.4ms (= 7071.9ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.03 (= 35.4ms / 34.4ms)

OneFlow swin dataloader time: 0.277s (= 55.401s / 200, num_workers=1)
PyTorch swin dataloader time: 0.154s (= 30.788s / 200, num_workers=1)
Relative speed: 0.556 (= 0.154s / 0.277s)

OneFlow swin dataloader time: 0.081s (= 16.198s / 200, num_workers=4)
PyTorch swin dataloader time: 0.041s (= 8.116s / 200, num_workers=4)
Relative speed: 0.501 (= 0.041s / 0.081s)

OneFlow swin dataloader time: 0.045s (= 9.016s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.462s / 200, num_workers=8)
Relative speed: 0.495 (= 0.022s / 0.045s)

❌ OneFlow resnet50 time: 144.8ms (= 14481.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 168.7ms (= 16866.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 168.7ms / 144.8ms)

OneFlow resnet50 time: 93.9ms (= 9386.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 112.4ms (= 11244.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 112.4ms / 93.9ms)

OneFlow resnet50 time: 70.7ms (= 14133.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 85.0ms (= 16995.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 85.0ms / 70.7ms)

OneFlow resnet50 time: 57.1ms (= 11421.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.0ms (= 15198.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 76.0ms / 57.1ms)

OneFlow resnet50 time: 51.7ms (= 10343.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 74.2ms (= 14842.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.43 (= 74.2ms / 51.7ms)