[hotfix] remove cuda half unittest in maxunpool

[marigoold]

max unpool 这个 op 的单测中，max_unpool3d 的 tensor 较大，会触发 https://github.com/Oneflow-Inc/OneTeam/issues/1783 中描述的问题，先暂时删掉 fp16 的测试，防止影响其他人的 PR 合并，等这个问题解决之后再修复

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 









❌ OneFlow resnet50 time: 140.2ms (= 14015.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.3ms (= 16027.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.14 (= 160.3ms / 140.2ms)

OneFlow resnet50 time: 85.3ms (= 8526.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 104.1ms (= 10405.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.22 (= 104.1ms / 85.3ms)

OneFlow resnet50 time: 58.0ms (= 11605.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.6ms (= 15722.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.35 (= 78.6ms / 58.0ms)

OneFlow resnet50 time: 44.3ms (= 8867.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 69.7ms (= 13936.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.57 (= 69.7ms / 44.3ms)

OneFlow resnet50 time: 40.2ms (= 8036.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 69.4ms (= 13882.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.73 (= 69.4ms / 40.2ms)

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9436/