Struct EpisodicMemorySchema 

pub struct EpisodicMemorySchema;

Marker struct for episodic agent memory tables (Phase 9). Actual schema is enforced by column names in episodic_columns module.

An episodic memory table extends LlmContextSchema with recency and importance signals, enabling hybrid scoring during recall:

-- From llm_columns::* (required baseline)
chunk_id:          Utf8
chunk_text:        Utf8
embedding:         FixedSizeBinary(N)  -- F16, cosine

-- From episodic_columns::* (Phase 9 extensions)
agent_id:          Utf8       -- UUID string
session_id:        Utf8       -- UUID string
created_at:        Timestamp(ns, UTC)  -- use ailake_core::now_ns()
recency_weight:    Float32    -- exp(-λ * days_since_access), updated by MemoryDecayJob
access_count:      UInt32     -- incremented on each recall() hit
last_accessed_at:  Timestamp(ns, UTC)  -- updated on recall, use ailake_core::now_ns()
importance_score:  Float32    -- agent-assigned [0.0, 1.0]

Hybrid scoring: after HNSW retrieval, re-rank results by hybrid_score(distance, recency_weight, importance_score). Memories that are semantically similar AND recently accessed AND flagged important rank highest.

Recommended setup:

• One HNSW over embedding (text, cosine, dim=1536).
• Partition by agent_id via VectorStoragePolicy hidden partitioning.
• Run MemoryDecayJob daily to update recency_weight via compaction.