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/cpu_common.go

@@ -0,0 +1,21 @@
+package gpu
+
+import (
+	"log"
+
+	"golang.org/x/sys/cpu"
+)
+
+func GetCPUVariant() string {
+	if cpu.X86.HasAVX2 {
+		log.Printf("CPU has AVX2")
+		return "avx2"
+	}
+	if cpu.X86.HasAVX {
+		log.Printf("CPU has AVX")
+		return "avx"
+	}
+	log.Printf("CPU does not have vector extensions")
+	// else LCD
+	return ""
+}