sync : ggml (#2001)

talk-llama : sync llama.cpp

ci : build with CLBlast + ggml-opencl use GGML_API (#1576)

examples : Revert CMakeLists.txt for talk-llama (#1669)

cmake : target windows 8 or above for prefetchVirtualMemory in llama-talk (#1617)

cmake : fix talk-llama build

sync : ggml (backend v2, k-quants, CUDA opts, Metal opts, etc.) (#1422)

whisper : Metal and ggml-alloc support (#1270)

whisper : add GPU support via cuBLAS (#834)

talk-llama : try to fix windows build ..

talk-llama : add new example + sync ggml from llama.cpp (#664)