Taxable-interest rebase: per-row control choice and per-pair bridge retirement can disagree #1236

Follow-up from the #1229 review (executed repros: ~/PolicyEngine/_reviews/cd-taxable-interest-20261012/inhouse/rebase-logic/, scenarios s5, s6b/s6c, s6f, s9).

_rebase_stale_soi_taxable_interest_distributions picks the control for each row. Bridge retirement is decided per (measure, status). The two rules can disagree, so a rebased row can name a control whose national row was retired, or a share family can keep a second national target. Neither corner occurs on the pinned 51aa40f feed: it has no Table 1.4 interest fact, and its district file is stamped no later than Table 4.3.

Corner 1: retired bridge used as a control

The setup:

• An exact all-returns control (Table 1.4) exists at an older period than a row.
• The row skips it for the eligible bridged Table 4.3 control. test_stale_soi_taxable_interest_skips_ineligible_control_for_eligible_bridge, from sol finding 5, pins this.
• retired_bridge_owners meanwhile retires the Table 4.3 national row, because an all-returns control exists at some period.

The row then lands on a level that no compiled target carries. Its uprating_index_source_record_id names a spec that is not in the registry.

Cases:

• On main: Historic Table 2 rows do this when Table 1.4 interest is present only at an older year than HT2 (s6f).
• On Rebase congressional-district taxable interest onto Table 4.3 as shares of the district file's US row #1229: with Table 1.4 at TY2022, Table 4.3 at TY2023, HT2 at TY2022 and the district file stamped ty2023, the district rows land on the retired Table 4.3 level. Its states sum to 2.17x the only national target, Table 1.4 aged (s6b/s6c). On main the same feed gives 0.87x plus the district US row as a second national target.

Corner 2: stamp after every control

If a share family's period stamp is later than every control, control is None keeps its rows as levels and its national all-AGI row as a second national target. One case is the next district file under the same +1 stamp error as 22incd.csv (Chronicle#117).

This happens on main and on #1229 alike. The stamp correction (#1030) prevents it for files it restamps.

Design question

What should own the national concept when an exact control and a bridged control are live at different periods?

• Option A: one owner per (measure, status) for every share family, chosen once (for example, the latest control in any universe), with each row's period gate applied to that owner.
• Option B: retire a bridge surrogate only when no share row rebases onto it.

Either reverses part of #518's lattice: "exact universe outranks the bridge at any period" against "an ineligible exact control must not shadow an eligible bridge". So it needs an adjudication of the #489 control lattice, not a drive-by.

Acceptance:

• A unit test with Table 1.4 TY2022, Table 4.3 TY2023, HT2 TY2022 and a district package stamped ty2023 shows every interest family summing to the same compiled national target.
• Every rebased row's control is a compiled spec. Rebase congressional-district taxable interest onto Table 4.3 as shares of the district file's US row #1229's Hypothesis property already asserts this on the shapes it draws.