Since #108, `raw_results.columns` requires `input_data__flat` even when `processed_data` is supplied directly

[MImmesberger]

Bug

#108 (4351a45) added input_data__flat and input_data__sort_indices as unconditional parameters of raw_results.columns (needed to reverse-translate endogenous p_id_* columns to user-space p_ids).

This breaks the path where users bypass input processing and pass processed_data directly:

main(
    main_target=("raw_results", "columns"),
    policy_date=policy_date,
    specialized_environment=SpecializedEnvironment.with_partialled_params_and_scalars(env),
    processed_data=processed_data,
    tt_targets=TTTargets.qname([qname]),
    backend=backend,
)

now fails the interface-DAG root check:

ValueError: The following arguments to `main` are missing for computing the desired output:

[
    "('input_data', 'flat')",
]

Before #108, providing processed_data pruned the input_data branch of the interface DAG entirely.

Impact

All 506 parametrized cases of GETTSIM's test_jittability.py fail with ttsim-backend 1.2.2.dev11+gad3d21c12 (e.g. https://github.com/ttsim-dev/gettsim/actions/runs/27062743460/job/79878306974).

Resolution

Split raw_results.columns into a symmetric pair (#131):

• raw_results.columns_with_internal_p_ids — computable from processed_data alone; endogenous p_id_* columns hold internal indices.
• raw_results.columns_with_original_p_ids — the Reverse-translate endogenous p_id_* targets to user p_ids (closes #43) #108 reverse-translation; consumed by results.tree, so all results.* outputs keep returning user-space p_ids.

The plain raw_results.columns target is removed; GETTSIM jittability tests switch to columns_with_internal_p_ids in ttsim-dev/gettsim#1196.

[hmgaudecker]

Ah, sorry, only saw this now. I guess this has to go in first before the next GETTSIM PR's or should we just ignore it for the day?

[MImmesberger]

I'm already working on the PR but I think we can merge the GETTSIM PRs nonetheless.