Add templates for recip_to_real in the pw_basis

[A-006]

What's changed?

• Currently, pw_basis utilizes the recip_to_real function directly, which does not support multi-device environments. To address this, I have added function templates to enable multi-device usage.
• Given that real2recip and recip2real are widely used throughout the program, transitioning to a multi-device setup requires specifying the context (ctx) for each. This change necessitates modifications in over 50 places within the program, most of which will require passing additional parameters. To simplify this transition, we have set the default context as CPU.
• For GPU and DCU functions of real_to_recip, it is necessary to pass pointers allocated in video memory (GPU). Therefore, memory allocation on the GPU must precede its use. Currently, rho is allocated on the CPU; hence, while we will add GPU functions, they will not be utilized until rho is also moved to GPU memory.