Threat diagnosis with OpenAI reasoning published to SOG

[CodeBuildder]

Context

OpenAI OpenAI model produces structured threat diagnosis for high/critical findings. Parent: #84.

Acceptance Criteria

• OpenAI model invoked for findings with severity >= high
• Diagnosis output: threat_type, attack_stage (MITRE ATT&CK phase), recommended_response, confidence
• Diagnosis written to Sentinel Operations Graph (SOG) as correlated finding referencing original finding_id
• Reasoning trace stored and queryable from Sentinel Operations Graph (SOG)

Dependencies

Blocked by: none | Blocks: none

Notes

Model tier: OpenAI model.