Offset SSTB and non-SSTB Schedule C results in federal AGI

[PavelMakarchuk]

Fixes #9989

What and why

irs_gross_income counts each Schedule C category's profit on its own (self_employment_income and sstb_self_employment_income are separate entries in gov.irs.gross_income.sources, each floored at zero), but loss_ald took the loss only from the net total_self_employment_income. A loss in one category next to a profit in the other was dropped: a non-SSTB loss of 5,000 and an SSTB profit of 5,000 gave AGI 5,000.

loss_ald now takes each category's profit and loss on its own, the same way irs_gross_income counts them. Per the 2024 Schedule C (lines 31 and 32a), both results go on Schedule 1 line 3, which flows to Form 1040 lines 8, 9 and 11, so AGI is 0. #9989 has the verbatim form quotes.

• § 461(l): the cap compares the same totals. With business profit I, loss L and threshold T, the allowed net is I − min(L, I + T) = max(I − L, −T) whether or not the categories are netted first.
• Self-employment tax already nets the two categories and does not change.
• Not changed: dc_self_employment_loss_addition also uses the net of the two categories. D.C.'s addition is a separate rule, so this PR leaves it as is.
• Open PRs Give each spouse their own above-the-line deductions in per-person AGI #9890 and Count capital gains once in Arizona property tax credit income #9762 move this block into limited_business_loss.py with the same two lines. Whichever merges after this one needs the same change there.

Test plan

• Three YAML tests in self_employment_income.yaml. Expected values come from the 2024 Schedule C, Schedule 1 and Form 1040 arithmetic in Federal AGI drops a Schedule C loss when the same person has Schedule C profit in the other SSTB category #9989 (AGI 0, 0 and 50,000). All three fail on main (07227ea252) and pass with this change.
• IRS YAML shard (test_batched.py policyengine_us/tests/policy/baseline/gov/irs --mode per-subdir --workers 1): 1,692 passed, 0 failed (local macOS run)
• Other YAML files that reference loss_ald or sstb_self_employment_income (49 files): 361 passed, 0 failed
• ruff format and ruff check (ruff 0.15.5, as in uv.lock)
• Changelog fragment in changelog.d/

axiom: TheAxiomFoundation/rulespec-us#1542 queued (§62(a)(1) composition, deferred in us/statutes/26/62.yaml; the signed encoder run needs approval)

🤖 Generated with Claude Code

[PavelMakarchuk]

Fixed the two red checks:

• Full Suite - Rest: tests/core/test_dependent_losses_agi_invariants.py::test_units_without_dependents_match_all_member_formula (2024, 2025) failed because its all-member reference netted self_employment_income and sstb_self_employment_income per person before splitting income from loss. With this PR loss_ald takes each Schedule C category on its own, as irs_gross_income does, so the reference now uses each input on its own. The nine Python test files that reference loss_ald or sstb_self_employment_income pass (78).
• Full Suite - Baseline (states-shard-3): the Mississippi HSA test that was red on main at the time; main is merged in (it was fixed by Update the Mississippi HSA test for joint-column combination (main is red) #9992).