llama.cpp Release b11557 Improves RPC Debugging and Logging

llama.cpp release b11557 introduces a tiered verbosity level for GGML_RPC_DEBUG and unifies logging for both clients and servers through a shared log.h header. The update adds missing instrumentation for handshakes, buffer operations, tensor transfers, and graph computes across the remote procedure call backend. This release significantly improves the observability and diagnostic capabilities of distributed and remote model inference in llama.cpp. Developers can now more easily troubleshoot network bottlenecks, performance issues, and communication errors between RPC clients and servers. GGML_RPC_DEBUG is now parsed as a numerical level ranging from 0 to 3 to control output detail, while degraded operations like RDMA falling back to TCP are promoted to warnings. Note that standard llama.cpp applications suppress these debug records unless the global verbosity threshold is explicitly raised using -lv 5.