[part2]Remove singleclient outdated api

[doombeaker]

移除 single client 下的不必要的接口 part 2。
等 part1合并后，这个分支直接往 master 合并。

[jackalcooper]

为什么把summary去掉了？

[chengtbf]

watch暂时不要删吧 @strint @clackhan @leaves-zwx @doombeaker ，lazy 新版的 nn.Graph 可能还用得到

[chengtbf]

也可以后面再加回来，因为没有下面的 hob 判断了（global_function 和 enable_eager）

[doombeaker]

这套代码 compatible_single_client_python 里还有一个。我不了解正在开发的 nn.Graph。不过我的朴素理解是，refine完的效果是：

1. 原来静态图那套（watch这种），都在 compatible_single_client_python 里
2. interface 1.0 的这套，都在 python 目录下

至于 interface 1.0 的接口和老接口有交集，就交给新接口的开发者自己去搬比较好……

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 137.6ms (= 6880.4ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 123.9ms (= 6196.1ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 137.6ms / 123.9ms)

PyTorch resnet50 time: 82.2ms (= 4109.9ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 72.0ms (= 3601.2ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 82.2ms / 72.0ms)

PyTorch resnet50 time: 56.1ms (= 2802.6ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 56.9ms (= 2847.2ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 0.98 (= 56.1ms / 56.9ms)

PyTorch resnet50 time: 47.5ms (= 2376.6ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 50.8ms (= 2540.0ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 0.94 (= 47.5ms / 50.8ms)

PyTorch resnet50 time: 41.1ms (= 2055.9ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 50.4ms (= 2520.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.82 (= 41.1ms / 50.4ms)