[Feature] Add Batch FFT for CUDA

[Critsium-xy]

According to tests from 沐曦's engineers, using batch fft on GPU to replace some FFT loop may make caculation faster in some cases. This PR added a input parameter which allows users to choose whether to use batch fft.

[Flying-dragon-boxing]

I'm not sure if the batch size should be a input parameter or merely a function parameter that can be differently decided when developers use batch FFT to do different things. As for me, this parameter might be a situation-specific one.

[Critsium-xy]

I'm not sure if the batch size should be a input parameter or merely a function parameter that can be differently decided when developers use batch FFT to do different things. As for me, this parameter might be a situation-specific one.

I've added it as a function parameter in FFT module. But according to previous test, setting batch_size to different size leads to different caculation speed. I cannot easily figure out which size is the best so maybe leave this problem to user is an option😤

[Critsium-xy]

I'm not sure when to continue this PR, so I close this for now.