sync : llama.cpp (fused soft max, gpu cpy ops, etc.) (#640)

sync : llama.cpp (fused soft max, gpu cpy ops, etc.) (#640)

* sync : llama.cpp (fused soft max, gpu cpy ops, etc.)

ggml-ci

* cuda : restore accidentally deleted changes

ggml-ci

* cuda : fix rope + disable device-side dequantize

ggml-ci

* test-backend-ops : enable stablelm rope test

* cuda : remove rope assert

* sync.sh : add test-backend-ops

* ggml : fix ggml_concat + ggml_get_n_tasks logic

* sync : whisper.cpp

ggml-ci

* metal : fix assert

* ci : fix Metal path to shaders

ggml-ci

* whisper : fix bug if metal init fails

---------

Co-authored-by: slaren <redacted>