Optimize FFT performance using batched execution with configurable batch size

[wangtianxiang]

This PR introduces performance optimization for FFT computations by leveraging batched FFT execution (via batchFFT). To provide flexibility and backward compatibility, a new parameter fft_batch_size is added with the following behavior:

• fft_batch_size = 0 (default): Automatically determine the optimal FFT batch size based on system and problem characteristics.
• fft_batch_size = 1: Disable batching and execute FFTs sequentially (identical to the original behavior).
• fft_batch_size > 1: Use batched FFT with a maximum batch size of fft_batch_size.

Note that on NVIDIA GPUs, batched FFT can be slower than looped FFT in many cases; it is recommended to set fft_batch_size = 1 to disable batching and retain the original behavior.

Signed-off-by：Tianxiang Wang[email protected],Contributed under MetaX Integrated Circuits (Shanghai) Co., Ltd.