chore: ⬆️ Update ggml-org/llama.cpp to 74ce15741b420b8d6f12e720398458b576c51c2c

[localai-bot]

Changes: https://github.com/ggml-org/llama.cpp/compare/221f0f6356efe2260023208365705ec5d5a7c8f5..74ce15741b420b8d6f12e720398458b576c51c2c

[mudler]

@localai-org-maint-bot fix this

[localai-org-maint-bot]

Fixed and pushed in 3b1d05109. The llama.cpp gRPC wrapper now calls the new eval_llama_cmpl_schema signature in both completion paths while preserving the legacy params_from_json_cmpl fallback. The completed lint, Linux/Apple tests, llama.cpp smoke, Darwin backend build, and current CPU/Vulkan backend jobs are green; the remaining backend matrix jobs are still running.

[mudler]

@localai-org-maint-bot check again failures

[mudler]

@localai-org-maint-bot fix CI issues

[localai-org-maint-bot]

Picked this back up and pushed a53e703c8 on top of current master. The pinned llama.cpp revision removed slot_n_ctx from eval_llama_cmpl_schema but the legacy fallback still requires it, so the modern and legacy calls now have separate preprocessor branches with their exact signatures. I verified both call sites against upstream 74ce1574 and git diff --check; the long backend builds are left to the newly triggered CI run.