Fix argwhere bug

[BBuf]

解决下面的问题：

[leaves-zwx]

这里的处理我想了下，也会有问题，argwhere 应该也不能支持 input 为 consistent，因为这个计算求出来的 res 里面的 nd-index 一定是 local 的，size 也是。所以改成：

if input.is_consistent:
    raise ValueError("A consistent tensor can not be applied to argwhere, and use `tensor.to_local()` to convert it to local tensor first.")

(res, size) = flow.F.argwhere(input, dtype=dtype)
if input.is_lazy:
    raise NotImplementedError
    # return flow.F.sync_dynamic_resize(res, size, dim=0)
else:
    slice_tup_list = [(0, size.numpy().item(), 1)]
    return flow.slice(res, slice_tup_list=slice_tup_list)

[BBuf]

好的

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 140.5ms (= 7024.9ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.2ms (= 6409.6ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 140.5ms / 128.2ms)

PyTorch resnet50 time: 84.2ms (= 4210.3ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.4ms (= 3720.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 84.2ms / 74.4ms)

PyTorch resnet50 time: 61.3ms (= 3062.6ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 51.2ms (= 2558.1ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.20 (= 61.3ms / 51.2ms)

PyTorch resnet50 time: 48.3ms (= 2415.7ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 40.4ms (= 2022.0ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.19 (= 48.3ms / 40.4ms)

PyTorch resnet50 time: 43.6ms (= 2180.0ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 43.5ms (= 2174.9ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.00 (= 43.6ms / 43.5ms)