refactor(sofiia-console): extract idempotency store abstraction

Move idempotency TTL/LRU logic into a dedicated store module with a swap-ready interface and wire chat send flow to use store get/set semantics without changing API behavior.

Made-with: Cursor

services/sofiia-console/app/idempotency.py

@@ -0,0 +1,89 @@
+from __future__ import annotations
+
+import os
+import time
+from collections import OrderedDict
+from dataclasses import dataclass
+from typing import Any, Dict, Optional, Protocol
+
+
+@dataclass
+class ReplayEntry:
+    message_id: str
+    response_body: Dict[str, Any]
+    created_at: float
+    node_id: str
+
+
+class IdempotencyStore(Protocol):
+    def get(self, key: str) -> Optional[ReplayEntry]:
+        ...
+
+    def set(self, key: str, entry: ReplayEntry) -> None:
+        ...
+
+
+class InMemoryIdempotencyStore:
+    def __init__(self, ttl_seconds: int = 900, max_size: int = 5000) -> None:
+        self._ttl_seconds = max(60, int(ttl_seconds))
+        self._max_size = max(100, int(max_size))
+        self._values: "OrderedDict[str, Dict[str, Any]]" = OrderedDict()
+
+    def _cleanup(self, now: Optional[float] = None) -> None:
+        ts = now if now is not None else time.monotonic()
+        while self._values:
+            first_key = next(iter(self._values))
+            exp = float((self._values[first_key] or {}).get("expires_at", 0.0))
+            if exp > ts:
+                break
+            self._values.popitem(last=False)
+
+    def get(self, key: str) -> Optional[ReplayEntry]:
+        self._cleanup()
+        hit = self._values.get(key)
+        if not hit:
+            return None
+        # Touch key to preserve LRU behavior.
+        self._values.move_to_end(key, last=True)
+        entry = hit.get("entry")
+        return entry if isinstance(entry, ReplayEntry) else None
+
+    def set(self, key: str, entry: ReplayEntry) -> None:
+        now = time.monotonic()
+        self._cleanup(now)
+        self._values[key] = {
+            "expires_at": now + self._ttl_seconds,
+            "entry": entry,
+        }
+        self._values.move_to_end(key, last=True)
+        while len(self._values) > self._max_size:
+            self._values.popitem(last=False)
+
+    # Debug/test helpers
+    def size(self) -> int:
+        self._cleanup()
+        return len(self._values)
+
+    def delete(self, key: str) -> None:
+        self._values.pop(key, None)
+
+    def reset(self) -> None:
+        self._values.clear()
+
+
+_STORE: Optional[IdempotencyStore] = None
+
+
+def get_idempotency_store() -> IdempotencyStore:
+    global _STORE
+    if _STORE is None:
+        ttl = int(
+            os.getenv(
+                "SOFIIA_IDEMPOTENCY_TTL_S",
+                os.getenv("CHAT_IDEMPOTENCY_TTL_SEC", "900"),
+            )
+        )
+        max_size = int(os.getenv("SOFIIA_IDEMPOTENCY_MAX", "5000"))
+        _STORE = InMemoryIdempotencyStore(ttl_seconds=ttl, max_size=max_size)
+    return _STORE
+