fix(transfer): remap whole-bank mental model evidence

[PaleRoses]

Problem

Whole-bank import restores mental-model rows twice: once through the embedded document archive before regenerated fact IDs are known, then again through the bank-level restore after remapping. The second insert loses to ON CONFLICT, leaving current evidence pointed at source-bank fact UUIDs. Historical evidence also stores its prior reflect response inside mental_model_history.content, which the existing remapper does not descend into.

This is a follow-up to #3833.

Change

• Skip document-level mental-model/page restoration for whole-bank archives; the bank importer remains the sole owner of those rows.
• Remap content.previous_reflect_response.based_on in mental-model history alongside current evidence.
• Preserve document-archive behavior unchanged.

Verification

Before the patch:

test_bank_roundtrip_remaps_mental_model_based_on_ids: FAILED
restored_ids contained the deleted source fact UUID

After the patch:

PYTHONPATH=<worktree>/hindsight-api-slim .venv/bin/pytest -q \
  hindsight-api-slim/tests/test_document_transfer.py::test_bank_roundtrip_remaps_mental_model_based_on_ids
1 passed

.venv/bin/ruff check hindsight-api-slim/hindsight_api/engine/transfer/importer.py
All checks passed!

[strix-security]

Strix Security Review

No security issues found.

_{Updated for 62829df.}

---

Reviewed by Strix
_{Re-run review · Configure security review settings}