First stable release marking the completion of Phase 0 through Phase 7.16
of the implementation plan. Every major subsystem is shipped and verified
in the slim Docker image.

Highlights:

- Slim Docker image (2 GB, down from 14 GB original MiroFish)
- Multi-provider LLM pool with probe, seeded allocation, capability
  detection, and automatic JSON-mode fallback for providers without
  response_format support
- Per-step LLM routing (ontology, profiles, config, simulation, report)
- Token tracking with pre-flight estimation, cost warnings, and
  pool-wide 24h budget aggregation
- Multi-model persona assignment (agents randomly routed to different
  providers; agent locked to assigned model for run duration)
- Full task cancellation with batch-boundary checkpoints
- Retry from failed/cancelled builds without losing progress
- Document quality scoring, CSV stakeholder input, orphan cleanup,
  subprocess reconciliation after server restart
- Internationalisation in 7 languages (en, zh, es, de, fr, pt, it)
- Pre-flight modal with 6-tier cost estimates and provider pool status
- Research-backed personality frameworks:
    * Individuals: Big Five / FFM (Costa & McCrae, John & Srivastava)
    * Institutions: 8 archetypes + 5 behavioural traits (informed by
      Mintzberg, Miles & Snow)
    * MBTI retained as legacy display field
- Weekly GitHub Actions audit of provider catalogue drift
- Comprehensive docs: ARCHITECTURE.md, oasis_dev.md,
  personality_frameworks.md, llm_providers.md, llm_budget_planning.md,
  document_and_ontology_guide.md, implementation_plan.md,
  new_features_planning.md