Disable GL_KHR_cooperative_matrix Vulkan extension if not available

[pepijndevos]

fixes #2733

[ggerganov]

This is synced now on master.

[pepijndevos]

Huh do you have some automated system to synchronise changes from llama.cpp?

[ggerganov]

It's semi-automated - running Bash scripts manually: ggml-org/ggml#1066

[pepijndevos]

Could I have another sync for ggml-org/llama.cpp#11406 to fix a segfault in ggml-org/llama.cpp#11308

Alternatively, do you accept PRs for synchronization or you'd rather just run it yourself?

[ggerganov]

Will sync today. Ideally, I would like to change the repositories so that no synchronization is required and all of them would use a single ggml source. But I haven't figured out the best way to do it yet.

[ggerganov]

There is something weird going on with the CI and the builds are failing - #2779. Will check later today if the issue disappears.

[slaren]

May be the same issue that we had with the ARM docker images:
ggml-org/llama.cpp#11434 (comment)
docker/build-push-action#1309