optimize of_include_copy

[daquexian]

1. 把拷贝第三方库头文件的操作移动到 prepare_oneflow_third_party 里，使得 -DTHIRD_PARTY=OFF 的时候不需要拷贝头文件
2. 把一些 add_custom_target（总是运行）改为 add_custom_command（在 depends 改变时才运行）

效果：
-DTHIRD_PARTY=OFF 时，在没有文件被修改的情况下运行 ninja oneflow_py 的时间从 36s 降低到了 1s，修改了单个 cpp 文件时的编译时间从 72s 降低到 46s

[daquexian]

先不合并，还有进一步的优化

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 140.7ms (= 7036.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.1ms (= 6403.4ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 140.7ms / 128.1ms)

PyTorch resnet50 time: 83.1ms (= 4154.8ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.3ms (= 3714.8ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 83.1ms / 74.3ms)

PyTorch resnet50 time: 54.1ms (= 2706.6ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.4ms (= 2372.3ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 54.1ms / 47.4ms)

PyTorch resnet50 time: 48.7ms (= 2435.0ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 38.8ms (= 1938.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.26 (= 48.7ms / 38.8ms)

PyTorch resnet50 time: 42.6ms (= 2129.5ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 36.5ms (= 1823.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.17 (= 42.6ms / 36.5ms)

[daquexian]

add_custom_target 总是 会运行，add_custom_command 只会在 DEPENDS 改变的时候运行

[daquexian]

不需要通过 target dependency 来触发命令，因为它的 OUTPUT（${${HDRS}} ${${SRCS}} ${${PYBIND_SRCS}}）会被某个 target 所包含，编译那个 target 的时候会自动执行命令

[github-actions]

CI failed, removing label automerge

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 140.2ms (= 7009.5ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.6ms (= 6428.5ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.09 (= 140.2ms / 128.6ms)

PyTorch resnet50 time: 84.3ms (= 4214.2ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.7ms (= 3736.0ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 84.3ms / 74.7ms)

PyTorch resnet50 time: 59.1ms (= 2956.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.8ms (= 2392.5ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.24 (= 59.1ms / 47.8ms)

PyTorch resnet50 time: 48.9ms (= 2446.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 41.3ms (= 2067.4ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.18 (= 48.9ms / 41.3ms)

PyTorch resnet50 time: 42.7ms (= 2133.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 37.3ms (= 1862.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.15 (= 42.7ms / 37.3ms)