Scalar::As() support being called in device

[EsdeathYZH]

No description provided.

[hjchen2]

要不直接把Maybe的接口去掉，default的时候assert

[liujuncheng]

要不直接把Maybe的接口去掉，default的时候assert

我觉得可以，这个类的构造方法就保证了初始化吧，所以应该不需要调用者检查错误

[EsdeathYZH]

要不直接把Maybe的接口去掉，default的时候assert

这样的话要改动的地方蛮多的，functional里面用Scalar.As很多

[liujuncheng]

要不直接把Maybe的接口去掉，default的时候assert

这样的话要改动的地方蛮多的，functional里面用Scalar.As很多

如果去人这个 Maybe 是没必要，多改几个地方也没关系，因为这个改动应该不会导致额外的错误， 或者修改 As 接口直接返回 Maybe::Ok(Value())，这样总是返回ok，可以减少其他地方的改动，保持兼容，后面新的调用鼓励用Value

[EsdeathYZH]

要不直接把Maybe的接口去掉，default的时候assert

这样的话要改动的地方蛮多的，functional里面用Scalar.As很多

如果去人这个 Maybe 是没必要，多改几个地方也没关系，因为这个改动应该不会导致额外的错误， 或者修改 As 接口直接返回 Maybe::Ok(Value())，这样总是返回ok，可以减少其他地方的改动，保持兼容，后面新的调用鼓励用Value

那就还是都改掉吧，我先把涉及到的地方都改掉你们再确认下maybe是不是可以直接移除的

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8348/

[github-actions]

Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally.

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 130.1ms (= 13008.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 144.1ms (= 14414.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 144.1ms / 130.1ms)

OneFlow resnet50 time: 76.4ms (= 7642.9ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 87.1ms (= 8709.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.14 (= 87.1ms / 76.4ms)

OneFlow resnet50 time: 54.0ms (= 10807.3ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 54.6ms (= 10928.6ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.01 (= 54.6ms / 54.0ms)

OneFlow resnet50 time: 42.5ms (= 8495.9ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 40.0ms (= 8009.1ms / 200, input_shape=[2, 3, 224, 224])
❌ Relative speed: 0.94 (= 40.0ms / 42.5ms)

OneFlow resnet50 time: 37.8ms (= 7566.6ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 37.2ms (= 7441.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 0.98 (= 37.2ms / 37.8ms)

OneFlow swin dataloader time: 0.245s (= 48.957s / 200, num_workers=1)
PyTorch swin dataloader time: 0.149s (= 29.827s / 200, num_workers=1)
Relative speed: 0.609 (= 0.149s / 0.245s)

OneFlow swin dataloader time: 0.104s (= 20.861s / 200, num_workers=4)
PyTorch swin dataloader time: 0.041s (= 8.183s / 200, num_workers=4)
Relative speed: 0.392 (= 0.041s / 0.104s)

OneFlow swin dataloader time: 0.035s (= 7.095s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.411s / 200, num_workers=8)
Relative speed: 0.622 (= 0.022s / 0.035s)

❌ OneFlow resnet50 time: 146.7ms (= 14667.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 175.0ms (= 17498.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.19 (= 175.0ms / 146.7ms)

OneFlow resnet50 time: 97.2ms (= 9717.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 113.3ms (= 11330.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.17 (= 113.3ms / 97.2ms)

OneFlow resnet50 time: 73.1ms (= 14622.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 87.5ms (= 17507.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 87.5ms / 73.1ms)

OneFlow resnet50 time: 61.4ms (= 12281.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 75.3ms (= 15054.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.23 (= 75.3ms / 61.4ms)

OneFlow resnet50 time: 55.3ms (= 11061.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 73.8ms (= 14754.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 73.8ms / 55.3ms)