Add Intelligence Tasks

[mariano]

This adds the foundation for Intelligence Tasks: authorable task definitions, deterministic canonicalization, privacy-safe validation, direct in-memory execution, agent target integration, sanitized events and usage, file and artifact handling, retention hooks, and a minimal CLI validation command.

Roadmap

• Phase 0: Decision Lock
• Phase 1: Definition And Validation Foundation
• Phase 2: In-Memory Store And Direct Agent Execution
• Phase 3: Files, Artifacts, Conversion, And Retention Hooks
• Phase 4: PostgreSQL Store, Queue, Worker, And Idempotency
• Phase 5: Observability And Provider Usage Hardening
• Phase 6: CLI And SDK Product Completion
• Phase 7: Conditional Flow Support
• Phase 8: Production Hardening