fix return type error on gcc 4.8.5

[daquexian]

通过在 UNIMPLEMENTED() 后面加 return、把 UNIMPLEMENTED() 改成 UNIMPLEMENTED_THEN_RETURN() 消除编译错误，在 zhenhua 的环境下验证可行

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 140.6ms (= 7030.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 126.3ms (= 6314.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 140.6ms / 126.3ms)

PyTorch resnet50 time: 85.0ms (= 4249.3ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 73.8ms (= 3691.3ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.15 (= 85.0ms / 73.8ms)

PyTorch resnet50 time: 57.6ms (= 2879.1ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.2ms (= 2361.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.22 (= 57.6ms / 47.2ms)

PyTorch resnet50 time: 49.2ms (= 2460.8ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 42.0ms (= 2099.3ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.17 (= 49.2ms / 42.0ms)

PyTorch resnet50 time: 42.6ms (= 2131.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 42.0ms (= 2100.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.01 (= 42.6ms / 42.0ms)