llmosafe 0.5.2

Safety-critical cognitive safety library for AI agents. 4-tier architecture (Resource Body, Kernel, Working Memory, Sifter) with formal verification primitives, detection layer, and integration primitives.