fix: correct timestamp parsing bug - chained .replace() was stripping timezone

The chained .replace("Z", "+00:00").replace("+00:00", "") calls were
undoing each other, causing Z-suffixed timestamps to lose timezone info.
Now strips "Z" directly and ensures naive datetime for cutoff comparison.
Added regression test for old Z-suffixed timestamps.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/utils.py

@@ -64,7 +64,7 @@ def filter_memories_by_time(memories: List[Dict], hours: int = 24) -> List[Dict]
             try:
                 # Parse ISO timestamp
                 if isinstance(ts, str):
-                    mem_time = datetime.fromisoformat(ts.replace("Z", "+00:00").replace("+00:00", ""))
+                    mem_time = datetime.fromisoformat(ts.replace("Z", "")).replace(tzinfo=None)
                 else:
                     mem_time = ts
                 if mem_time > cutoff:

tests/test_utils.py

@@ -124,6 +124,16 @@ class TestFilterMemoriesByTime:
         result = filter_memories_by_time(memories, hours=24)
         assert len(result) == 1
 
+    def test_z_suffix_old_timestamp_excluded(self):
+        """Regression: chained .replace() was not properly handling Z suffix on old timestamps."""
+        from datetime import datetime, timedelta, timezone
+        from app.utils import filter_memories_by_time
+
+        old_ts = (datetime.now(timezone.utc).replace(tzinfo=None) - timedelta(hours=48)).isoformat() + "Z"
+        memories = [{"timestamp": old_ts, "text": "old with Z"}]
+        result = filter_memories_by_time(memories, hours=24)
+        assert len(result) == 0, f"Old Z-suffixed timestamp should be excluded but wasn't: {old_ts}"
+
     def test_empty_list(self):
         """Empty input returns empty list."""
         from app.utils import filter_memories_by_time