Use __sync_synchronize() hardware barrier instead of asm compiler barrier

sdk/software/examples/asm/user-sample.c

@@ -206,7 +206,7 @@ int main(void)
     }
 
     /* ── CRC32 over result region ─────────────────────────────── */
-    __asm__ __volatile__("" ::: "memory");  // flush write buffer
+    __sync_synchronize();  // hardware data barrier — drain write buffer
     crc = 0xFFFFFFFF;
     for (i = 0; i < RESULT_WORDS; i++) {
         uint32_t w = extram[DST_BASE + i];