Update LLM01_PromptInjection.md with Agentic AI hazards, wormability, and timeline

Key Changes:

• Broadened prompt injection definition to include untrusted data and tool call results
• Added memory compromise follow-on attack vector
• Expanded impacts to cover wormable and persistent attacks
• Added historical context for early prompt injection disclosure

Added:

• Scenario for persistent access via poisoned LLM memory
• Scenario for prompt virus and worm-style propagation
• Timeline and future risks section
• References for early disclosure, public rediscovery, and mitigation

Changed:

• Reordered impact bullets to emphasize risks involving agentic systems
• Updated description to cover behavior, reasoning, tools, and actions
• Expanded reference list with prompt injection history sources