llama.cpp b10447

server: re-design yield_to_queue thread model (#27133)

• run common_speculative_process in worker

• swap worker main thread design

Website: -

UI:

• UI