Always dynamically load the llm server library

This switches darwin to dynamic loading, and refactors the code now that no
static linking of the library is used on any platform

gpu/gpu_darwin.go

@@ -32,8 +32,15 @@ func CheckVRAM() (int64, error) {
 
 func GetGPUInfo() GpuInfo {
 	mem, _ := getCPUMem()
+	if runtime.GOARCH == "amd64" {
+		return GpuInfo{
+			Library: "default",
+			Variant: GetCPUVariant(),
+			memInfo: mem,
+		}
+	}
 	return GpuInfo{
-		Library: "default",
+		Library: "metal",
 		memInfo: mem,
 	}
 }
@@ -45,12 +52,3 @@ func getCPUMem() (memInfo, error) {
 		DeviceCount: 0,
 	}, nil
 }
-
-func nativeInit() error {
-	return nil
-}
-
-func GetCPUVariant() string {
-	// We don't yet have CPU based builds for Darwin...
-	return ""
-}