fix(turboquant): repair static wht declaration

The bumped fork adds an explicit extern after GGML_API, but GGML_API already expands to extern in static builds. Carry a compatibility patch that keeps the symbol declaration-only without producing conflicting specifiers.

Assisted-by: Codex:gpt-5 [Codex]

Author: localai-org-maint-bot

backend/cpp/turboquant/patches/0002-fix-static-wht-group-size-declaration.patch

@@ -0,0 +1,20 @@
+ggml-cpu: avoid duplicate extern in static builds
+
+GGML_API expands to extern when GGML_SHARED is disabled. Combining it with
+an explicit extern therefore produces an invalid `extern extern` declaration
+in the static builds used by LocalAI. The definition in ggml-turbo-quant.c
+already carries GGML_API, so this cross-library declaration does not need the
+export macro.
+
+diff --git a/ggml/src/ggml-cpu/ops.cpp b/ggml/src/ggml-cpu/ops.cpp
+index 755e9ff8c..2a624b90c 100644
+--- a/ggml/src/ggml-cpu/ops.cpp
++++ b/ggml/src/ggml-cpu/ops.cpp
+@@ -18,6 +18,6 @@ extern "C" {
+ // handler here and the quantizer there operate on different variables. Whether
+ // the two happen to unify is a property of the platform's symbol resolution
+ // (ELF interposition may merge them; two-level-namespace and DLL targets will
+ // not), which made the group-size propagation silently link-order dependent.
+-GGML_API extern int turbo3_cpu_wht_group_size;
++extern int turbo3_cpu_wht_group_size;
+ }