fix(memory): only delete memories the model explicitly drops in tidy (#3455)

* fix(memory): only delete memories the model explicitly drops in tidy

The AI memory-tidy path computed deletions as the complement of the model's
`keep` list (`if mid not in keep_ids: continue`). When the model returned a
valid response that simply omitted some existing ids — a common LLM lapse — every
omitted memory was silently deleted, even though it was neither a duplicate nor
listed in `drop`.

Honor the explicit `drop` set instead: delete only ids the model dropped (minus
any it saw only truncated), and preserve everything else, still applying cleaned
text/category from `keep`.

Adds tests/test_consolidate_memory_explicit_drops.py: a memory the model omits
from both keep and drop survives; an explicitly dropped one is removed.

* refactor(memory): remove now-dead keep_ids from tidy

After deletion switched to drop_ids and text/category rewrites to cleaned_by_id,
keep_ids was written but never read. Remove the init, the .add(mid) in the keep
loop, and the truncated .update() (its truncated-protection is already covered by
`drop_ids -= truncated_ids`). Pure deletion, no behavior change; tests stay green.

Addresses review feedback on #3455.

---------

Co-authored-by: Kenny Van de Maele <kenny@kvandemaele.be>

tests/test_consolidate_memory_explicit_drops.py

@@ -0,0 +1,57 @@
+"""Memory consolidation must delete only memories the model explicitly drops.
+
+The AI tidy path computed deletions as the complement of the model's `keep`
+list, so any memory the model simply omitted (a common LLM lapse) was silently
+deleted. The fix honors the explicit `drop` set, so an omitted memory survives.
+"""
+import asyncio
+import json
+
+import src.builtin_actions as ba
+
+
+class _FakeMM:
+    saved = None
+
+    def __init__(self, *args, **kwargs):
+        pass
+
+    def load_all(self):
+        return [
+            {"id": "a", "owner": "alice", "text": "Likes dark roast coffee", "category": "preference"},
+            {"id": "b", "owner": "alice", "text": "Likes dark roast coffee too", "category": "preference"},
+            {"id": "c", "owner": "alice", "text": "Lives in Cairo", "category": "fact"},
+        ]
+
+    def save(self, entries):
+        _FakeMM.saved = list(entries)
+
+
+def test_omitted_memory_survives_only_explicit_drop(monkeypatch):
+    import src.memory
+    import src.endpoint_resolver
+    import src.llm_core
+
+    _FakeMM.saved = None
+    monkeypatch.setattr(src.memory, "MemoryManager", _FakeMM)
+    monkeypatch.setattr(
+        src.endpoint_resolver, "resolve_endpoint",
+        lambda kind, owner=None: ("http://x/v1", "model", {}),
+    )
+
+    async def fake_llm(**kwargs):
+        # Model keeps 'a', drops 'b', and OMITS 'c' entirely.
+        return json.dumps({
+            "keep": [{"id": "a", "text": "Likes dark roast coffee", "category": "preference"}],
+            "drop": [{"id": "b", "reason": "duplicate of a"}],
+        })
+
+    monkeypatch.setattr(src.llm_core, "llm_call_async", fake_llm)
+
+    msg, ok = asyncio.run(ba.action_consolidate_memory("alice"))
+
+    assert ok, msg
+    ids = {m["id"] for m in _FakeMM.saved}
+    assert "c" in ids, "omitted memory must NOT be deleted"
+    assert "a" in ids
+    assert "b" not in ids, "explicitly dropped memory should be removed"