Add unimplemented return information

[luqiang-guo]

优化之后的出错信息提示如下：

例如执行

import oneflow as flow

shape = (2, 3)
placement = flow.placement("cuda", {0: 0})
sbp = flow.sbp.split(0)
a = flow.Tensor(*shape)
b = a.to_consistent(placement=placement, sbp=sbp)

b.device()

修改前报错信息如下：

Traceback (most recent call last):
  File "test.py", line 9, in <module>
    b.device()
oneflow._oneflow_internal.exception.UnimplementedException: 
  File "/home/guoluqiang/oneflow_tmp/oneflow-1/oneflow/core/framework/tensor.h", line 439, in device

修改之后的错误信息如下：

Traceback (most recent call last):
  File "test.py", line 9, in <module>
    b.device()
oneflow._oneflow_internal.exception.RuntimeException: 
  File "/home/guoluqiang/oneflow_tmp/oneflow/oneflow/core/framework/tensor.h", line 492, in device
    RuntimeError : ConsistentTensor has no device property

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.6ms (= 6978.6ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.0ms (= 6398.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.09 (= 139.6ms / 128.0ms)

PyTorch resnet50 time: 84.3ms (= 4214.8ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.8ms (= 3740.0ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 84.3ms / 74.8ms)

PyTorch resnet50 time: 58.3ms (= 2912.5ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 48.0ms (= 2397.5ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.21 (= 58.3ms / 48.0ms)

PyTorch resnet50 time: 47.7ms (= 2386.3ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 43.3ms (= 2166.3ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 47.7ms / 43.3ms)

PyTorch resnet50 time: 39.1ms (= 1956.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 41.5ms (= 2075.4ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.94 (= 39.1ms / 41.5ms)

[daquexian]

目前的错误信息感觉还有改进的空间，它应该面向用户而不是内部开发者。写错误信息的时候应该考虑用户会在这么场景下触发这个错误，然后提示给用户尽可能丰富的信息。

比如 tensor 的 parallel_desc 方法，目前的错误信息是 "MirroredTensor has no parallel_desc property"，但 "MirroredTensor" 和 "parallel_desc" 都是用户陌生的概念（因为它们没有暴露到 python 层），而且错误信息也没有告诉用户该怎么做。parallel_desc 方法对应 python api 里的 placement 方法，更好一些的错误信息应该是 "Only consistent tensors have 'placement', please use '.device()' for local tensors." （placement 和 device 两侧的引号对提高用户可读性很有帮助）.

再比如 AsConsistentTensor 方法，"MirroredTensor has no AsConsistentTensor property" 这个信息对用户来说也有疏离感，改成 "An error occured when converting a local tensor to consistent tensor. Check if you are calling consistent methods on local tensors" 就好一些

我先去掉了 automerge label，如果决定在下一个 pr 里再修复这些问题可以再添加回来

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 141.4ms (= 7067.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.5ms (= 6425.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 141.4ms / 128.5ms)

PyTorch resnet50 time: 85.1ms (= 4252.9ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.8ms (= 3738.5ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 85.1ms / 74.8ms)

PyTorch resnet50 time: 55.3ms (= 2764.1ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 50.1ms (= 2502.6ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 55.3ms / 50.1ms)

PyTorch resnet50 time: 44.9ms (= 2242.8ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 42.9ms (= 2147.1ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.04 (= 44.9ms / 42.9ms)

PyTorch resnet50 time: 43.2ms (= 2159.5ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 38.4ms (= 1919.8ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 43.2ms / 38.4ms)