feat: implement DCOS source tree, tests, and build configuration

Complete implementation of the DCOS package including:
- 40+ Python source files across agents, communication, core, learning, memory, protocols, utils
- pyproject.toml build configuration
- 102 unit tests across all subsystems
- Fixed flake.nix (Python 3.12, proper dependencies, pyproject build)
- Fixed .woodpecker.yml (correct paths, removed silent-fail flags)
- Added .gitignore
- Cleared stale pytest cache

src/dcos/learning/engine.py

@@ -0,0 +1,44 @@
+"""
+Learning engine — reinforcement learning and model updates for agents.
+"""
+
+from __future__ import annotations
+from typing import Any, Dict, List, Optional
+from dataclasses import dataclass, field
+
+
+@dataclass
+class Experience:
+    state: Dict[str, Any]
+    action: str
+    reward: float
+    next_state: Optional[Dict[str, Any]] = None
+
+
+class LearningEngine:
+    """Reinforcement learning engine that improves agent behavior over time."""
+
+    def __init__(self) -> None:
+        self._experiences: List[Experience] = []
+        self._policy: Dict[str, float] = {}
+
+    def record_experience(self, exp: Experience) -> None:
+        self._experiences.append(exp)
+        self._update_policy(exp)
+
+    def _update_policy(self, exp: Experience) -> None:
+        current = self._policy.get(exp.action, 0.0)
+        self._policy[exp.action] = current + exp.reward * 0.1
+
+    def get_action_score(self, action: str) -> float:
+        return self._policy.get(action, 0.0)
+
+    def best_action(self, actions: List[str]) -> Optional[str]:
+        if not actions:
+            return None
+        scored = [(a, self._policy.get(a, 0.0)) for a in actions]
+        scored.sort(key=lambda x: x[1], reverse=True)
+        return scored[0][0]
+
+    def experience_count(self) -> int:
+        return len(self._experiences)