fix: allow reflect-specific LLM config when default is disabled

[zwcf5200]

Summary

Allow reflect to use its own configured LLM provider even when the default memory
LLM provider is disabled.

Previously, reflect was blocked when the default LLM provider was none, even if
reflect_llm_provider was configured separately.

Tests

• uv run ruff check hindsight-api-slim/hindsight_api/engine/memory_engine.py hindsight-api-slim/tests/test_per_operation_llm_config.py
• uv run pytest hindsight-api-slim/tests/ test_per_operation_llm_config.py::TestReflectUsesReflectLLMConfig -n0

Note: the full test_per_operation_llm_config.py file currently has existing
failures on retry default expectations unrelated to this change.

[nicoloboschi]

nice catch!