Systematically Analyzing Prompt Injection Vulnerabilities in Diverse LLM Architectures