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/learning/pattern_miner.py

@@ -0,0 +1,42 @@
+from __future__ import annotations
+
+from uuid import uuid4
+
+from app.agents.schemas.learning import LearningSignal, PatternCandidate
+
+
+class LearningPatternMiner:
+    def mine(self, signals: list[LearningSignal]) -> list[PatternCandidate]:
+        patterns: list[PatternCandidate] = []
+
+        for signal in signals:
+            if signal.signal_type not in {"workflow", "decomposition", "preference"}:
+                continue
+
+            description = self._build_description(signal)
+            patterns.append(
+                PatternCandidate(
+                    pattern_id=f"pattern-{uuid4().hex[:10]}",
+                    pattern_type=signal.signal_type,
+                    description=description,
+                    confidence=signal.confidence,
+                    evidence_refs=signal.evidence_refs[:4],
+                )
+            )
+
+        return patterns
+
+    @staticmethod
+    def _build_description(signal: LearningSignal) -> str:
+        payload = signal.payload or {}
+        if signal.signal_type == "workflow":
+            task_type = payload.get("task_type") or "general"
+            execution_mode = payload.get("execution_mode") or "direct"
+            return f"Completed {task_type} requests worked under {execution_mode} execution."
+        if signal.signal_type == "decomposition":
+            task_count = payload.get("task_count") or 0
+            return f"Requests with {task_count} concrete task refs benefit from structured decomposition."
+        if signal.signal_type == "preference":
+            preference = payload.get("preference") or "structured response"
+            return f"User preference repeatedly points to {preference}."
+        return signal.explanation or signal.signal_type