feat(memory): episodic memory consolidation pipeline

[Mathews-Tom]

Summary

Add a memory consolidation pipeline for episodic memory lifecycle management, inspired by memora-lab/memory-service-public's observation lifecycle patterns. Episodes now have a complete lifecycle: created → resolved → archived → summarized into lesson notes → promoted to permanent vault notes. This is the final PR (#49) in the memora-inspired enhancement plan.

Three-Step Consolidation

MemoryConsolidator.consolidate()
  ├─ Step 1: Archive old resolved episodes
  │    └─ Episodes resolved > 365 days ago → outcome_status = "archived"
  │
  ├─ Step 2: Summarize archived clusters
  │    ├─ Group archived episodes by shared entities
  │    ├─ Clusters with 3+ episodes → LLM synthesis
  │    └─ Write lesson notes to _meta/lessons/{entity}-{date}.md
  │
  └─ Step 3: Promote frequently-referenced episodes
       ├─ Find resolved episodes whose entities appear in 3+ other episodes
       └─ Write permanent notes to _meta/episodes/{decision}-{date}.md

Episode Lifecycle (Complete)

PENDING → SUCCESS/FAILURE/PARTIAL/UNKNOWN → ARCHIVED → lesson note / promoted note
   ↑              ↑                            ↑              ↑
 create()      resolve()          archive_old_resolved()  consolidate()

Changes

New Files

• src/vaultmind/memory/consolidation.py (241 lines) — MemoryConsolidator class with consolidate() orchestrator, _archive_old_episodes(), _summarize_clusters() (groups by shared entity, LLM synthesis, writes lesson notes), _promote_referenced() (cross-reference counting, writes permanent notes). ConsolidationReport dataclass tracks counts and errors
• tests/test_consolidation.py (230 lines) — 14 tests across 5 classes

Modified Files

• src/vaultmind/memory/models.py — Added ARCHIVED = "archived" to OutcomeStatus enum, enabling the archive lifecycle state
• src/vaultmind/memory/store.py — Three new methods:
  • archive_old_resolved(age_days=365) — UPDATE resolved episodes older than threshold to "archived" status. Skips pending and already-archived. Returns count
  • query_archived(limit=100) — retrieve archived episodes ordered by created desc
  • count_entity_references(entity) — count episodes mentioning a given entity (for promotion threshold)
• src/vaultmind/config.py — Added ConsolidationConfig class with enabled (default False), retention_days (365), min_references_for_promotion (3), schedule ("0 0 1 * *" — monthly). Added consolidation field to Settings
• config/default.toml — Added [consolidation] section with all config entries

Backward Compatibility

• ConsolidationConfig.enabled defaults to False — zero behavior change for existing users
• ARCHIVED status is additive to OutcomeStatus enum — existing statuses unaffected
• New store methods are additive — existing query/resolve/create methods unchanged
• All existing episodic tests pass unchanged (17 in test_episodic.py, 15 in test_memory_horizon.py)
• Consolidation writes to _meta/lessons/ and _meta/episodes/ — isolated from user content

Output Examples

Lesson note (_meta/lessons/lesson-kubernetes-20260326.md):

---
title: Lessons — kubernetes
type: permanent
tags: [lesson, consolidation]
---
# Lessons — kubernetes
- Lesson 1: Always test
- Lesson 2: Keep it simple
---
*Synthesized from 4 episodes.*

Promoted note (_meta/episodes/episode-api-decision-20260326.md):

---
title: API decision 0
type: permanent
tags: [episode, promoted]
---
# API decision 0
**Context:** ...
**Outcome:** Result 0 (success)
## Lessons
- lesson 0
**Entities:** api

Test plan

• 14 new tests in test_consolidation.py across 5 classes:
  • Archive: old resolved archived, recent skipped, pending skipped, already-archived skipped (4)
  • Query archived: returns archived, empty when none (2)
  • Entity references: counts correctly across episodes (1)
  • Consolidator: archives, summarizes clusters with LLM, skips without LLM, promotes referenced, disabled no-op (5)
  • Config: defaults, Settings integration (2)
• All existing episodic/horizon tests pass unchanged
• Full suite: 1028/1028 tests pass, 0 regressions
• ruff check — clean
• mypy --ignore-missing-imports — clean
• Integration: wire into scheduler as monthly job
• Manual: run consolidation on vault with 365+ day old episodes, verify lesson notes