Share one HHS SMI helper for LIHEAP income limits

[hua7450]

Fixes #10005

What and why

Mississippi, Maryland and New Jersey LIHEAP each computed the HHS 60% state median income (SMI) limit with their own copy of the household-size factor and rounding. Colorado CCAP repeated the factor too. This PR moves those calculations into policyengine_us/variables/gov/hhs/hhs_smi.py:

• smi_size_factor(unit_size, period, parameters) returns the 45 CFR 96.85(b) share of the four-person SMI. smi() now calls it.
• liheap_smi_limit(unit_size, state, rate, period, parameters) floors the rate times the four-person SMI, multiplies by the size factor, and floors again. HHS describes the multiplication in its footnote (LIHEAP IM 2025-02, Attachment 4) but gives no rounding order; this floor order reproduces the published tables. Callers that need a prior year's SMI pass that date as period.

Caller	Change
ms_liheap_income_limit	Calls liheap_smi_limit; keeps its size > 0 guard
md_meap_income_limit	Calls liheap_smi_limit for sizes 11+
nj_liheap_income_limit	Calls liheap_smi_limit for each table size; keeps its half-up monthly rounding and per-member step
co_ccap_smi	Calls smi() at its October date, then divides by 12

NJ and MD merged after #10005 was filed, so the issue does not list them; they ran the same floor-then-floor calculation. Georgia (#9968), Florida (#9969) and Ohio (#9975) switch to the helper in their own PRs.

Behavior

The PR adds no parameters and changes no test expectations.

• MS and CO keep the same arithmetic. CO previously hard-coded 6 for the size cutoff; it now reads additional_person_threshold (6), which matches for every unit with at least one person.
• MD and NJ previously derived the size factor as smi() / four-person SMI. The helper uses the factor directly. A comparison over every state, every SMI date in amount.yaml and sizes 1-20 (14,540 cases) found no NJ or MD difference. The direct factor matched exact rational arithmetic in all 14,540 cases. The division route was $1 low in 9 other-state cases (for example IL FFY2022 size 10) from floating-point error.

Test plan

• YAML tests from an external source: not applicable; no expected value changes. CI runs the existing MS, MD, NJ, CO CCAP and smi() consumer tests.
• ruff format and ruff check on the changed files
• Changelog fragment in changelog.d/
• CI passes

axiom: n/a: refactor with no policy or parameter change

🤖 Generated with Claude Code