fix batch gather get_sbp and add global unittest

[hjchen2]

No description provided.

[wyg1997]

我看 batch_gather 的文档中的示例还是用的 flow.Tensor 接口，后江可以顺手改成 flow.tensor 接口嘛

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.6ms (= 12856.4ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 139.1ms (= 13911.9ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.08 (= 139.1ms / 128.6ms)

✔️ OneFlow resnet50 time: 76.8ms (= 7678.0ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.2ms (= 8421.2ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.10 (= 84.2ms / 76.8ms)

OneFlow resnet50 time: 52.8ms (= 10569.7ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 58.7ms (= 11730.4ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.11 (= 58.7ms / 52.8ms)

OneFlow resnet50 time: 42.4ms (= 8481.0ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 48.7ms (= 9749.0ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.15 (= 48.7ms / 42.4ms)

OneFlow resnet50 time: 39.7ms (= 7933.8ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.5ms (= 8100.5ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.02 (= 40.5ms / 39.7ms)

✔️ OneFlow resnet50 time: 142.5ms (= 14253.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.6ms (= 16062.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.13 (= 160.6ms / 142.5ms)

OneFlow resnet50 time: 91.3ms (= 9129.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 101.9ms (= 10188.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 101.9ms / 91.3ms)

OneFlow resnet50 time: 63.5ms (= 12693.5ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.2ms (= 15244.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 76.2ms / 63.5ms)

OneFlow resnet50 time: 53.9ms (= 10786.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 72.4ms (= 14473.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.34 (= 72.4ms / 53.9ms)

OneFlow resnet50 time: 49.3ms (= 9851.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.5ms (= 12901.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 64.5ms / 49.3ms)

[hjchen2]

我看 batch_gather 的文档中的示例还是用的 flow.Tensor 接口，后江可以顺手改成 flow.tensor 接口嘛

以后再改吧，和这个PR也没有太大联系，而且这个PR已经在跑CI了，更新了又要重新跑了