brpc 支持 GPU 直接 RDMA

Is your feature request related to a problem?

Describe the solution you'd like When training on the GPU, use RDMA transport with brpc, and expect to use GDR capabilities to directly write the RPC request to the GPU's buffer, and then parse the protocol part by copying it from the GPU using cudamemcp or gdrcopy, in order to reduce data transfer between the CPU and GPU.