Enhance KomaMRI.jl GPU Support

KomaMRI.jl is a Julia package for Magnetic Resonance Imaging (MRI) simulation focused on speed and extensibility. Like other MRI simulators, it solves the Bloch equations, for which the solution can be parallelized due to the independent spin property of the system. Although KomaMRI currently has built-in support for GPU and CPU parallelization, GPU support is CUDA-based and could be extended to work with different GPU backends. In addition, writing GPU kernels tailored for the solution implementation could significantly improve performance compared with the current array-based approach. The goals for this project are to update the KomaMRI.jl package to support Metal, AMD, and OneAPI-based GPU backends, and create a new BlochKernel simulation method which is fully kernel-based and optimized. If time permits, the use of MPI.jl and Distributed.jl to run MRI simulations across multiple nodes and GPUs will also be explored.