[PERFORMANCE]: Fix critical performance issues in llm-guard plugin

[araujof]

Tasks:

• Async Redis operations - Single biggest bottleneck
• Parallelize scanner execution - 3-5x speedup potential
• Offload ML inference to threads - Prevent event loop blocking
• Fix pickle blocking - Use async for large objects