feat(learning): add learning runtime with pattern mining

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

backend/app/agents/schemas/learning.py

@@ -0,0 +1,76 @@
+from __future__ import annotations
+
+from datetime import datetime, timezone
+from typing import Any, Literal
+
+from pydantic import BaseModel, Field
+
+
+LearningSignalType = Literal[
+    "preference",
+    "workflow",
+    "decomposition",
+    "tool_success",
+    "correction",
+]
+
+
+class SessionRetrospective(BaseModel):
+    retrospective_id: str | None = None
+    user_id: str
+    conversation_id: str
+    request_message_id: str | None = None
+    response_message_id: str | None = None
+    query_text: str
+    final_response: str | None = None
+    summary: str
+    task_type: str | None = None
+    execution_mode: str | None = None
+    primary_agent: str | None = None
+    verification_status: str | None = None
+    verification_summary: str | None = None
+    used_skill_names: list[str] = Field(default_factory=list)
+    evidence_refs: list[dict[str, Any]] = Field(default_factory=list)
+    task_refs: list[dict[str, Any]] = Field(default_factory=list)
+    event_refs: list[dict[str, Any]] = Field(default_factory=list)
+    context_snapshot: dict[str, Any] = Field(default_factory=dict)
+    learning_signals: list["LearningSignal"] = Field(default_factory=list)
+    pattern_candidates: list["PatternCandidate"] = Field(default_factory=list)
+    skill_candidates: list["SkillCandidate"] = Field(default_factory=list)
+    learning_decision: "LearningDecision | None" = None
+    outcome: Literal["completed", "partial", "failed"] = "completed"
+    captured_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
+
+
+class LearningSignal(BaseModel):
+    signal_type: LearningSignalType
+    confidence: float = 0.0
+    evidence_refs: list[dict[str, Any]] = Field(default_factory=list)
+    explanation: str | None = None
+    payload: dict[str, Any] = Field(default_factory=dict)
+
+
+class PatternCandidate(BaseModel):
+    pattern_id: str
+    pattern_type: str
+    description: str
+    confidence: float = 0.0
+    evidence_refs: list[dict[str, Any]] = Field(default_factory=list)
+
+
+class SkillCandidate(BaseModel):
+    candidate_id: str
+    name: str
+    summary: str
+    candidate_type: Literal["workflow_skill", "preference_skill", "decomposition_skill"] = "workflow_skill"
+    source_pattern_ids: list[str] = Field(default_factory=list)
+    confidence: float = 0.0
+    evidence_refs: list[dict[str, Any]] = Field(default_factory=list)
+    recommended_status: Literal["candidate", "shadow"] = "candidate"
+
+
+class LearningDecision(BaseModel):
+    decision: Literal["reinforce_memory", "create_candidate", "promote_skill", "defer", "reject"]
+    explanation: str
+    evidence_refs: list[dict[str, Any]] = Field(default_factory=list)
+    metadata: dict[str, Any] = Field(default_factory=dict)