GCB 2025 emissions, IGCC 2025 budgets with peak-warming bands, LULUCF v4, start years from 1850; budget rebase fixes

[setupelz]

Summary

This PR adds three opt-in data sources and one deduction rule, lifts three limits on early start years, and corrects three accounting errors in the budget rebase. Defaults stay as they are (PRIMAP emissions, melo-2026 LULUCF, the existing budget sources and their AR6 category mapping).

Results change for existing budget sources. Adjusted budgets for sources with a baseline after 2020 rise. ar6_2020 and every allocation share are unchanged.

Source, label	co2-ffi before	after	co2 before	after
lamboll_2023 1.5p50	321,980	324,802	219,612	241,477
forster_2024 1.5p50	308,717	312,677	202,075	230,905
ar6_2020 (all labels)	unchanged		unchanged

Values are Mt CO2 from 2020, PRIMAP with melo-2026.

Decisions for the reviewer

1. Accept the two rebase corrections below as results-changing fixes in the next release.
2. Accept the peak-warming band rule for the two new budget sources.
3. Decide whether rcbs.yaml should become a declared pipeline input (follow-up 1).

Added

• gcb-2025 emissions source. Global Carbon Budget 2025 national fossil CO2, 1850 to 2024 (Andrew and Peters flat file, doi:10.5281/zenodo.17417124; Friedlingstein et al. 2026, doi:10.5194/essd-18-3211-2026). co2-ffi only. Blank years count as zero. The world row excludes international bunkers and the 1991 Kuwaiti oil fires. Bunkers come from the same file.
• Coverage rule for gcb-2025. A country is an analysis country only if the source records emissions for it before 1990 and its population series starts in 1850. Nine countries join rest-of-world. The rule is a coverage: block on the source; a source without the block behaves as before.
• forster_2026 and ar6_wg1_2021 budget sources (Indicators of Global Climate Change 2025, doi:10.5194/essd-18-3889-2026, Table 8; AR6 WGI Table SPM.2, doi:10.1017/9781009157896.001) for 1.5, 1.7 and 2°C.
• Peak-warming band rule. For the two new sources, the scenario set behind each deduction is every AR6 scenario whose median peak warming lies in [T - 0.05, T + 0.05) and whose CO2 reaches net zero. Band sizes are 14, 111 and 37 scenarios. One band serves every likelihood of its temperature. An unmapped label or an empty band raises.
• Placeholder fill. When the rebase needs a year after the last observed year of a world series, that year takes the last observed value, for at most rebase_fill_max_years years (default 1), with a warning each time.
• melo-2026-v4 LULUCF source. NGHGI LULUCF Data Hub v4.0.0, 2000 to 2024 (doi:10.5281/zenodo.22828743).
• gcb-2025 with melo-2026-v4 for co2 runs end to end. The LULUCF step now runs before the scenario step.

Changed

• Allocation years from 1850 are accepted (was 1900).
• With capability_reference_year set, the budget approaches require GDP only at that year.
• The responsibility-window check for co2 follows the first year of the frame that is passed.

Fixed

• Bunkers in the rebase. For a baseline after 2020 the rebase added fossil emissions without bunkers and the deduction started in 2020, so bunkers for 2020 to the baseline were deducted and never added. Effect: +2,822 Mt (lamboll_2023), +3,959 Mt (forster_2024).
• Convention gap for co2. The rebase adds observed inventory LULUCF for 2020 to the baseline, and the gap also converted bookkeeping to inventory from 2020, so those years were converted twice. The gap now runs from the baseline year (Weber et al. 2026, doi:10.1038/s41467-026-69078-9, Methods, Eqs. 2 and 3). Effect: about +19 to +26 Gt for the two post-2020 sources.
• The world co2 series no longer subtracts bunkers from a fossil series that already excludes them.
• Bunker deductions use the last observed bunker year in the data.
• Year-availability checks test values, so a blank-padded series raises.
• A circular import when fair_shares.library.config.models is the first import.
• compose_config re-runs when the packaged data-source table changes.
• Documentation: the income-floor price base (Baer 2013, doi:10.1002/wcc.201), the GDP price base (constant 2021 international dollars), and the description of the co2 rebase.

Tests

821 passed, 16 deselected on the branch head. The first two commits introduce test files in their final form, so those two commits are not green on their own.

tests/fixtures/python_api/ is refreshed in its own commit. The earlier fixtures predated the August Gini-membership and GDP update.

Follow-ups, not in this PR

1. rcbs.yaml is read by the RCB notebook and is not a declared input of any pipeline rule, so an edit to it does not trigger a re-run.
2. The pathway approaches still require GDP at the first allocation year when a capability reference year is set.
3. The bookkeeping side of the convention gap is the scenario AFOLU|Direct flux and the gap is a median across scenarios. Weber et al. use an observed bookkeeping-model mean for 2020 to 2023 and a mean across scenarios. The docs state both differences.