feat(llama-cpp-localai-paged): paged KV cache llama.cpp backend + cross-request prefix sharing + GB10 decode optimization [WIP]

feat(llama-cpp-localai-paged): paged KV cache llama.cpp backend + cross-request prefix sharing + GB10 decode optimization [WIP] #24144