[Tests] Extend semantic_memory_smoke with edge case coverage

## Summary
The semantic memory smoke tests need additional edge case coverage.

## Current State
- `semantic_memory_smoke.py`: 3 basic tests
- `semantic_memory_smoke_extended.py`: 17 extended tests
- Tests focus on "happy path" scenarios

## Missing Edge Cases
1. **Empty memory file handling**: Query with empty `.mem` file
2. **Corrupted memory file**: Handle truncated or invalid embedding rows
3. **Unicode/UTF-8 edge cases**: Facts with multi-byte characters
4. **Very long facts**: Facts exceeding typical token limits
5. **Concurrent access**: Multiple processes accessing same memory file
6. **Memory file permissions**: Read-only filesystem scenarios

## Acceptance Criteria
- [ ] Add `test_empty_memory_query` - verify graceful handling
- [ ] Add `test_corrupted_memory_recovery` - validate error messages
- [ ] Add `test_unicode_fact_storage` - UTF-8 roundtrip
- [ ] Add `test_very_long_fact_truncation` - document behavior
- [ ] Document expected behavior for each edge case

## Related Files
- `tools/longmemeval-mini/semantic_memory_smoke.py`
- `tools/longmemeval-mini/semantic_memory_smoke_extended.py`