fix: detect loader-accepted duplicate safetensors keys

feat: qualify Ollama and vLLM runtime topologies

Repair fresh Daybreak release blockers

Authenticate executable AI runtime evidence

Preserve fair discovery across recovery

Close Daybreak containment and deleted descriptor gaps

Inspect mapped model artifacts through proc descriptors

Repair Daybreak adversarial discovery invariants

Bound deep inspection cost and publish overhead benchmark

Repair release correctness and operational safety gates

Add aggressive Safetensors PyTorch containment profile

perf: defer runtime inspection until model evidence

fix: record content qualification timing

feat: add name-independent GGUF recognition