fix tensor method docstr bug

[BBuf]

修复了一部分Tensor方法没有docstr的bug。

• sqrt
• square
• addmm
• cosh
• diagonal
• log
• ndim
• rsqrt

剩下的：

• stride
• clone
• data
• detach
• dtype
• shape

是通过pybind11直接导出的，直接设置docstr会触发错误，后面需要和迟哥讨论一下再解决。

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/zhangxiaoyu/oneflow/python/oneflow/__init__.py", line 401, in <module>
    import oneflow.framework.docstr as docstr
  File "/home/zhangxiaoyu/oneflow/python/oneflow/framework/docstr/__init__.py", line 30, in <module>
    from .tensor import *
  File "/home/zhangxiaoyu/oneflow/python/oneflow/framework/docstr/tensor.py", line 1563, in <module>
    add_docstr(
  File "/home/zhangxiaoyu/oneflow/python/oneflow/framework/docstr/utils.py", line 32, in add_docstr
    return oneflow._oneflow_internal.add_doc(fun, docstr)
oneflow._oneflow_internal.exception.RuntimeException: 
  File "/home/zhangxiaoyu/oneflow/oneflow/api/python/framework/doc.cpp", line 56, in AddFunctionDoc
    RuntimeError: function is instancemethod, not a valid function.
>>> exit()

最后：norm这个算子的docstr比较复杂，设计到pytorch的新老接口替换的问题，也需要再讨论一下。

这个pr可以先合并，以下是本次补充的docstr的截图。

[github-actions]

CI failed when running job: cuda-module-distributed-rank-1. PR label automerge has been removed

[github-actions]

CI failed when running job: cuda-module-distributed-rank-0. PR label automerge has been removed

[doombeaker]

我发现，我用 add_docstr 是可以直接对 dtype 等内容添加文档的？

>>> import oneflow
>>> from oneflow.framework.docstr.utils import add_docstr
>>> print(oneflow.Tensor.dtype.__doc__)

>>> add_docstr(oneflow.Tensor.dtype, "it's ok")
<property object at 0x7f3d85bcb6b0>
>>> print(oneflow.Tensor.dtype.__doc__)
it's ok

oneflow 版本：

(oneflow-dev-gcc7-v2) yaochi@oneflow-23:~/build_exp$ python -m oneflow --doctor
path: ['/home/yaochi/oneflow/python/oneflow']
version: 0.8.0+cu112.git.1875c6e2b3
git_commit: 1875c6e2b3
cmake_build_type: Release
rdma: False
mlir: False

有点奇怪，要不要核对下是不是oneflow版本的问题？因为我有点记忆，后江修过类似的问题。我本来以为是还没考虑全情况，不过我这边测又没问题……

[github-actions]

CI failed when running job: cuda-module-distributed-rank-0. PR label automerge has been removed

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.4ms (= 12835.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.0ms (= 14101.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.10 (= 141.0ms / 128.4ms)

✔️ OneFlow resnet50 time: 77.3ms (= 7726.3ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.4ms (= 8443.4ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.09 (= 84.4ms / 77.3ms)

OneFlow resnet50 time: 52.2ms (= 10433.4ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 58.1ms (= 11610.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.11 (= 58.1ms / 52.2ms)

OneFlow resnet50 time: 42.7ms (= 8536.4ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.1ms (= 9816.3ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.15 (= 49.1ms / 42.7ms)

OneFlow resnet50 time: 38.3ms (= 7661.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 38.5ms (= 7691.3ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.00 (= 38.5ms / 38.3ms)

OneFlow swin dataloader time: 0.254s (= 50.788s / 200, num_workers=1)
PyTorch swin dataloader time: 0.260s (= 51.949s / 200, num_workers=1)
✔️ Relative speed: 1.023 (= 0.260s / 0.254s)

OneFlow swin dataloader time: 0.065s (= 12.961s / 200, num_workers=4)
PyTorch swin dataloader time: 0.070s (= 13.969s / 200, num_workers=4)
✔️ Relative speed: 1.078 (= 0.070s / 0.065s)

OneFlow swin dataloader time: 0.037s (= 7.436s / 200, num_workers=8)
PyTorch swin dataloader time: 0.038s (= 7.609s / 200, num_workers=8)
✔️ Relative speed: 1.023 (= 0.038s / 0.037s)

✔️ OneFlow resnet50 time: 135.9ms (= 13585.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 155.8ms (= 15576.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 155.8ms / 135.9ms)

OneFlow resnet50 time: 87.1ms (= 8710.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 108.0ms (= 10801.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 108.0ms / 87.1ms)

OneFlow resnet50 time: 61.0ms (= 12192.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.0ms (= 15192.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 76.0ms / 61.0ms)

OneFlow resnet50 time: 52.9ms (= 10589.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 67.9ms (= 13588.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.28 (= 67.9ms / 52.9ms)

OneFlow resnet50 time: 50.2ms (= 10043.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 71.7ms (= 14337.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.43 (= 71.7ms / 50.2ms)