docs(runbooks): add abductive inference and missing premise generation investigation playbook - #34

Review (blocking findings before merge):

1. Frontmatter violates the OKF v0.2 schema merged in Migrate frontmatter to OKF v0.2 conventions #32: uses type: Playbook + resource/timestamp/provenance, while all 32 runbooks use type: runbook, category, status, tags, generated {by, at}.
2. Missing every mandatory runbook section (Objective/Scope/Inputs/Tools/Rubric). Rubric appears in 32/32 existing runbooks and is absent here, breaking the eval-scoring pattern.
3. References nonexistent infrastructure: Neo4j/Cypher, AlloyDB, and a query_alloydb_detection_reports tool that exists nowhere in the repo or agent_tool_mapping.md. As written the playbook is unexecutable by any agent here. Either replace with real secops-mcp/gti-mcp tools or gate this PR behind feat(runbooks): add Neo4j graph traversal and AlloyDB vector similarity AI runbooks #33.
4. skills/abductive-inference/SKILL.md: name does not match directory, required_roles missing, personas use snake_case but manifests are kebab-case (resolve to nothing), no generated block, no Required Outputs section.
5. Citation misattributed: arXiv:2301.12987 (The Optimal Choice of Hypothesis Is the Weakest, Not the Shortest) is by Michael Timothy Bennett, not Evans/Stuhlmuller/Goodman; the SKILL.md entropy formulation also inverts the paper's claim.

Happy to merge once the frontmatter/sections/tool-grounding are aligned.