fix(feedback): handle empty added_ids in _single_add_operation

[lailoo]

Summary

• Bug: IndexError in _single_add_operation when memory_manager.add() returns empty list
• Root cause: feedback.py:254 accesses added_ids[0] without checking if the list is empty
• Fix: Add empty-list guard that returns {"id": None, ...} gracefully

Fixes #1122 (Part 2 of 2 — the IndexError on empty added_ids)

Problem

When the graph DB silently fails (e.g. due to the CypherTypeError in Part 1), _add_memories_parallel() catches the exception, logs it, but returns an empty list. Then _single_add_operation crashes:

added_ids = self._retry_db_operation(
    lambda: self.memory_manager.add([to_add_memory], user_name=user_name, use_batch=False)
)
return {
    "id": added_ids[0],  # IndexError: list index out of range
    ...
}

Before fix:

added_ids = []
added_ids[0]  ->  IndexError: list index out of range

Changes

• src/memos/mem_feedback/feedback.py — Add empty-list guard before accessing added_ids[0]
• tests/mem_feedback/test_feedback_empty_added_ids.py — Regression tests

After fix:

added_ids = []
-> returns {"id": None, "text": "...", "source_doc_id": None}  (no crash)

Test plan

• New test: test_single_add_returns_none_id_when_added_ids_empty — verifies graceful handling
• New test: test_single_add_returns_id_when_added_ids_present — verifies normal path

Effect on User Experience

Before: /product/feedback endpoint crashes with 500 Internal Server Error when the underlying graph DB add fails silently.
After: The endpoint returns a response with "id": null, allowing the caller to detect and handle the failure.