fix reduce_sum scalar check bug

[BBuf]

解决：https://github.com/Oneflow-Inc/OneTeam/issues/1229

已成功跑通，并且添加了单测。

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.3ms (= 12829.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 140.6ms (= 14056.7ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.10 (= 140.6ms / 128.3ms)

❌ OneFlow resnet50 time: 79.1ms (= 7914.3ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.9ms (= 8486.9ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.07 (= 84.9ms / 79.1ms)

OneFlow resnet50 time: 51.7ms (= 10334.7ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 56.3ms (= 11255.6ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.09 (= 56.3ms / 51.7ms)

OneFlow resnet50 time: 42.8ms (= 8560.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.6ms (= 9911.4ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.16 (= 49.6ms / 42.8ms)

OneFlow resnet50 time: 36.6ms (= 7317.0ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 38.8ms (= 7750.0ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.06 (= 38.8ms / 36.6ms)

md5 /home/ci-user/ci-cache/test_cache/onerec_test/nanodataset.zip 7f5cde8b5a6c411107517ac9b00f29db
md5 /home/ci-user/ci-cache/test_cache/onerec_test/nanodataset.zip 7f5cde8b5a6c411107517ac9b00f29db
build_dataset >>>>>> ImageFolder
build_dataset >>>>>> ImageFolder
✔️ Relative speed: 0.98 (= 51.1s / 50.3s)

md5 /home/ci-user/ci-cache/test_cache/onerec_test/nanodataset.zip 7f5cde8b5a6c411107517ac9b00f29db
md5 /home/ci-user/ci-cache/test_cache/onerec_test/nanodataset.zip 7f5cde8b5a6c411107517ac9b00f29db
build_dataset >>>>>> ImageFolder
build_dataset >>>>>> ImageFolder
✔️ Relative speed: 1.07 (= 13.6s / 14.6s)

md5 /home/ci-user/ci-cache/test_cache/onerec_test/nanodataset.zip 7f5cde8b5a6c411107517ac9b00f29db
md5 /home/ci-user/ci-cache/test_cache/onerec_test/nanodataset.zip 7f5cde8b5a6c411107517ac9b00f29db
build_dataset >>>>>> ImageFolder
build_dataset >>>>>> ImageFolder
✔️ Relative speed: 1.02 (= 7.5s / 7.7s)

✔️ OneFlow resnet50 time: 135.5ms (= 13553.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 162.0ms (= 16199.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 162.0ms / 135.5ms)

OneFlow resnet50 time: 87.6ms (= 8756.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 101.4ms (= 10139.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 101.4ms / 87.6ms)

OneFlow resnet50 time: 58.8ms (= 11750.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.4ms (= 15270.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.30 (= 76.4ms / 58.8ms)

OneFlow resnet50 time: 50.9ms (= 10183.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.6ms (= 13326.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 66.6ms / 50.9ms)

OneFlow resnet50 time: 47.3ms (= 9463.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 63.4ms (= 12672.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.34 (= 63.4ms / 47.3ms)

[github-actions]

CI failed when running job: cuda-speed-test. PR label automerge has been removed