MPI
每个 MPI 进程持有本地粒子和背景。DistributedMccStepper 先准备本地碰撞，协调错误状态和新产物 ID，再共同提交。所有相关进程都要参与调用；主程序保留全局唯一粒子 ID，负责 MPI 初始化、网格划分和粒子迁移。
mpi.hpp：分布式批量接口。
fun_G02_mcc_mpi.cpp：准备、协调与提交的实现。
构建时启用 G02_MCC_ENABLE_MPI，链接 AlgoPlasma::G02_MCC_MPI。MPI 线程支持至少为 MPI_THREAD_FUNNELED；在 MPI_Finalize() 前销毁步进器。
Each rank owns local particles/backgrounds. DistributedMccStepper prepares local collisions, agrees on errors and child IDs, then commits together. Every participating rank must call it. The host preserves globally unique IDs and owns MPI initialization, partitioning and particle migration.
mpi.hpp: distributed batch API.
fun_G02_mcc_mpi.cpp: preparation, agreement and commit.
Enable G02_MCC_ENABLE_MPI and link AlgoPlasma::G02_MCC_MPI. Require at least MPI_THREAD_FUNNELED and destroy the stepper before MPI_Finalize().