Fix select_first_grad bug

[wyg1997]

解决 oneflow-inc/oneteam#578 中 insight-face ddp 的问题。ddp 中使用的 select_first op ，加入的Parameters 列表可能有的 requires_grad 为 False，但仍会为其求出梯度，导致在 AutogradEngine 中检察报错。

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 125.6ms (= 6280.1ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 136.8ms (= 6837.6ms / 50, input_shape=[16, 3, 224, 224])
Relative speed: 1.09 (= 136.8ms / 125.6ms)

OneFlow resnet50 time: 73.2ms (= 3657.6ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 82.1ms (= 4106.8ms / 50, input_shape=[8, 3, 224, 224])
Relative speed: 1.12 (= 82.1ms / 73.2ms)

OneFlow resnet50 time: 47.5ms (= 2373.9ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.9ms (= 2897.0ms / 50, input_shape=[4, 3, 224, 224])
Relative speed: 1.22 (= 57.9ms / 47.5ms)

OneFlow resnet50 time: 46.2ms (= 2308.2ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.5ms (= 2474.4ms / 50, input_shape=[2, 3, 224, 224])
Relative speed: 1.07 (= 49.5ms / 46.2ms)

OneFlow resnet50 time: 41.5ms (= 2076.7ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.2ms (= 2009.2ms / 50, input_shape=[1, 3, 224, 224])
Relative speed: 0.97 (= 40.2ms / 41.5ms)

OneFlow resnet50 time: 141.4ms (= 7072.1ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 161.9ms (= 8095.4ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
Relative speed: 1.14 (= 161.9ms / 141.4ms)

OneFlow resnet50 time: 94.2ms (= 4711.9ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 104.5ms (= 5223.3ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
Relative speed: 1.11 (= 104.5ms / 94.2ms)

OneFlow resnet50 time: 71.2ms (= 3558.9ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.9ms (= 3843.5ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
Relative speed: 1.08 (= 76.9ms / 71.2ms)

OneFlow resnet50 time: 62.4ms (= 3119.1ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.7ms (= 3535.6ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
Relative speed: 1.13 (= 70.7ms / 62.4ms)

OneFlow resnet50 time: 61.0ms (= 3050.8ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.1ms (= 3205.8ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
Relative speed: 1.05 (= 64.1ms / 61.0ms)