koboldcpp v1.118: RPC compatibility and multi-GPU fixes

The latest release of koboldcpp (v1.118) introduces several critical updates for local AI operators running on consumer hardware. Notably, it aligns the RPC behavior with upstream standards, enabling seamless interoperability between llama.cpp RPC servers and clients alongside koboldcpp ones. Additionally, this version resolves a multi-GPU issue where incorrect GPU selection could occur, ensuring more reliable performance in multi-GPU setups. The update also adds support for qwen3tts aliases across all voices, enhancing compatibility with various voice models.