Add fused self attention

[BBuf]

添加 fused_self_attention_query_mul_key_and_value 为functor。

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 138.2ms (= 6908.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.3ms (= 6413.1ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.08 (= 138.2ms / 128.3ms)

PyTorch resnet50 time: 83.5ms (= 4177.1ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.5ms (= 3722.8ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 83.5ms / 74.5ms)

PyTorch resnet50 time: 55.0ms (= 2748.5ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.5ms (= 2375.4ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.16 (= 55.0ms / 47.5ms)

PyTorch resnet50 time: 49.9ms (= 2497.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 41.1ms (= 2055.7ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.21 (= 49.9ms / 41.1ms)

PyTorch resnet50 time: 42.9ms (= 2145.4ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 36.4ms (= 1821.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.18 (= 42.9ms / 36.4ms)