Record refresh mental model outcome metadata

[koriyoshi2041]

Summary

• add outcome metadata for completed async mental-model refresh operations
• record content length, populated-content signal, and based_on counts by fact type
• keep submit-time metadata such as mental_model_id/name intact

Closes #2605

Test

• uv run --with socksio pytest tests/test_mental_model_scheduled_refresh.py::test_async_refresh_writes_outcome_metadata -q -n0
• uv run ruff format --check hindsight_api/engine/memory_engine.py hindsight_api/engine/operation_metadata.py tests/test_mental_model_scheduled_refresh.py
• uv run ruff check hindsight_api/engine/memory_engine.py hindsight_api/engine/operation_metadata.py tests/test_mental_model_scheduled_refresh.py
• git diff --check

[koriyoshi2041]

Closing this in favor of #2627. That PR covers the same result_metadata parity gap and adds the no-answer stub boundary, which is a better fit for #2605 than keeping two overlapping implementations open.

I also checked #2627 locally with git diff --check and the focused refresh tests; see the validation comment there.