Remove CUDA-only restriction for multi-tensor model updates in optimizer

[ShawnXuan]

This PR removes the CUDA-only restriction in the multi-tensor model update pass within both multi_tensor_model_update.cpp and adamw.py.

By commenting out the device type checks, we allow multi-tensor and fused updates to run on non-CUDA devices as well.

[github-actions]

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10564/

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.7ms (= 4368.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.3ms (= 5731.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.31 (= 57.3ms / 43.7ms)

OneFlow resnet50 time: 26.1ms (= 2614.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.1ms (= 3714.5ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.42 (= 37.1ms / 26.1ms)

OneFlow resnet50 time: 19.0ms (= 3804.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 35.5ms (= 7103.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.87 (= 35.5ms / 19.0ms)

OneFlow resnet50 time: 17.5ms (= 3501.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 32.3ms (= 6456.6ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.84 (= 32.3ms / 17.5ms)

OneFlow resnet50 time: 17.4ms (= 3483.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 31.3ms (= 6253.9ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.80 (= 31.3ms / 17.4ms)

OneFlow swin dataloader time: 0.199s (= 39.839s / 200, num_workers=1)
PyTorch swin dataloader time: 0.129s (= 25.762s / 200, num_workers=1)
Relative speed: 0.647 (= 0.129s / 0.199s)

OneFlow swin dataloader time: 0.054s (= 10.725s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.542s / 200, num_workers=4)
Relative speed: 0.610 (= 0.033s / 0.054s)

OneFlow swin dataloader time: 0.030s (= 6.077s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.307s / 200, num_workers=8)
Relative speed: 0.544 (= 0.017s / 0.030s)

❌ OneFlow resnet50 time: 49.6ms (= 4961.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.0ms (= 6399.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.29 (= 64.0ms / 49.6ms)

OneFlow resnet50 time: 36.8ms (= 3680.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 45.6ms (= 4562.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 45.6ms / 36.8ms)

OneFlow resnet50 time: 28.1ms (= 5615.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.5ms (= 8098.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.44 (= 40.5ms / 28.1ms)

OneFlow resnet50 time: 25.1ms (= 5027.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.4ms (= 7687.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.53 (= 38.4ms / 25.1ms)

OneFlow resnet50 time: 24.9ms (= 4981.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 37.4ms (= 7476.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.50 (= 37.4ms / 24.9ms)

[Flowingsun007]

这个去掉，是不是cpu device会有影响，是否需要用param.is_cpu判断一下？

[github-actions]

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

[github-actions]

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