ollama: don't build llama.cpp twice

buildGoModules runs preBuild hooks in the module download fixed-output
derivation, so llama.cpp got built and immediately discarded, to then
immediately be built again for the actual Go build.

pkgs/by-name/ol/ollama/package.nix

@@ -181,6 +181,14 @@ goBuild (
       # replace inaccurate version number with actual release version
       substituteInPlace version/version.go --replace-fail 0.0.0 '${version}'
     '';
+
+    overrideModAttrs = (
+      finalAttrs: prevAttrs: {
+        # don't run llama.cpp build in the module fetch phase
+        preBuild = "";
+      }
+    );
+
     preBuild = ''
       # disable uses of `git`, since nix removes the git directory
       export OLLAMA_SKIP_PATCHING=true