Auto FixPipelineStageIdPass

[chengtbf]

用户配置 Pipeline stage id 并不一定可以严格的保证 stage 连续（尤其是 module 衔接处的 op，如 to_consistent），因此我们需要有个算法来自动修复出错的 stage id。（本身 用户配置的实际上也是 pipeline stage id hint，oneflow 内部有权修改为更合适的值）

算法：

1. 统计所有 placement 对应的 op name 和 stage id，对于同一个 Placement 下的所有 op，根据其 stage id 分组；
2. 如果一个 Placement 下有多组 stage id，那么将取 group 内 stage id 最大的为真正的 stage id，将剩余其他的 stage id 的 op 都 merge 到这个真正的 stage id 上。

[chengtbf]

重构了算法，现在主体逻辑大幅简化：

每个 Placement Group 寻找最大的 stage id，将其余的 op，如果 stage id 小于这个 最大值，就 merge 到最大值

@leaves-zwx @strint

自己本地测试是没问题的

[github-actions]

CI failed, removing label automerge

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 127.5ms (= 6376.5ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 139.2ms (= 6958.4ms / 50, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.09 (= 139.2ms / 127.5ms)

OneFlow resnet50 time: 74.2ms (= 3710.8ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.8ms (= 4191.8ms / 50, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.13 (= 83.8ms / 74.2ms)

OneFlow resnet50 time: 53.9ms (= 2697.0ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 59.0ms (= 2948.6ms / 50, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.09 (= 59.0ms / 53.9ms)

OneFlow resnet50 time: 44.6ms (= 2229.8ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 56.2ms (= 2811.7ms / 50, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.26 (= 56.2ms / 44.6ms)

OneFlow resnet50 time: 48.1ms (= 2406.7ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 47.3ms (= 2365.4ms / 50, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 0.98 (= 47.3ms / 48.1ms)

OneFlow resnet50 time: 155.5ms (= 7774.0ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 161.8ms (= 8089.5ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.04 (= 161.8ms / 155.5ms)

OneFlow resnet50 time: 99.5ms (= 4974.3ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 106.3ms (= 5313.2ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.07 (= 106.3ms / 99.5ms)

OneFlow resnet50 time: 79.6ms (= 3978.5ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 79.6ms (= 3981.0ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.00 (= 79.6ms / 79.6ms)

OneFlow resnet50 time: 72.8ms (= 3641.9ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.9ms (= 3546.1ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 0.97 (= 70.9ms / 72.8ms)

OneFlow resnet50 time: 71.0ms (= 3550.9ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.7ms (= 3286.4ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 0.93 (= 65.7ms / 71.0ms)

[github-actions]

CI failed, removing label automerge

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 127.6ms (= 6379.7ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.9ms (= 7093.5ms / 50, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 141.9ms / 127.6ms)

OneFlow resnet50 time: 74.4ms (= 3721.0ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 82.8ms (= 4138.6ms / 50, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.11 (= 82.8ms / 74.4ms)

OneFlow resnet50 time: 48.0ms (= 2400.7ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 56.7ms (= 2836.2ms / 50, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.18 (= 56.7ms / 48.0ms)

OneFlow resnet50 time: 43.1ms (= 2154.6ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 51.3ms (= 2567.0ms / 50, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.19 (= 51.3ms / 43.1ms)

OneFlow resnet50 time: 42.4ms (= 2121.7ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 39.1ms (= 1955.1ms / 50, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 0.92 (= 39.1ms / 42.4ms)

OneFlow resnet50 time: 153.0ms (= 7650.6ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 154.4ms (= 7717.9ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.01 (= 154.4ms / 153.0ms)

OneFlow resnet50 time: 99.3ms (= 4965.6ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 102.4ms (= 5117.8ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.03 (= 102.4ms / 99.3ms)

OneFlow resnet50 time: 78.3ms (= 3916.8ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 74.7ms (= 3736.6ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 0.95 (= 74.7ms / 78.3ms)

OneFlow resnet50 time: 73.1ms (= 3657.0ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.6ms (= 3278.7ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 0.90 (= 65.6ms / 73.1ms)

OneFlow resnet50 time: 72.3ms (= 3614.9ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.6ms (= 3232.4ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 0.89 (= 64.6ms / 72.3ms)