Exported input DataFrame no longer round-trips: main CI fails test_an_exported_frame_loads_back_and_exports_itself #2333

policyengine_uk/tests/test_dataframe_dataset.py::test_an_exported_frame_loads_back_and_exports_itself fails on main. It failed in the "Code changes" run on 7343f94 (run 38143054188) and fails locally on 41e970b:

DataFrame shape mismatch
[left]:  (1, 38)
[right]: (1, 18)

Repro. The falsifying example is one person, one benunit and one household in LONDON, with zero inputs:

exported = Simulation(dataset=frame(population, "both")).to_input_dataframe()
again = Simulation(dataset=exported).to_input_dataframe()

Cause. exported already holds uc_receives_new_claimant_health_element__2026 … __2029 next to the 2025 inputs. That variable came in with #2251, fix-uc-health-claim-status. Loading that frame back adds the entity id columns for 2026-2029: person_id, benunit_id, household_id, person_benunit_id and person_household_id, __2026 to __2029. So the second export has 20 more columns than the first, and the round-trip identity fails.

Fix options.

• Don't export values the simulation set for years other than the dataset year.
• Or export the id columns for every year that holds inputs, so the export is a fixed point.

The test's invariant (export ∘ load ∘ export = export) is the right one, so keep it.