Impute is_pregnant in CPS using CDC birth rates + add calibration targets

[daphnehanse11]

Problem

is_pregnant is a pure input variable in policyengine-us with no formula and no data in any CPS dataset. This means zero people are pregnant in microsimulations, breaking:

• Medicaid pregnancy category enrollment
• WIC pregnancy eligibility and household size
• CHIP (standard + FCEP pathways)
• SNAP ABAWD exemptions
• Medicaid work requirement exemptions
• State TANF programs (MT, IL, DC, CO, MA, AK, WI)
• Various state Medicaid programs (CA immigration exemption, CO OmniSalud, etc.)

The CPS does not ask about current pregnancy status. A previous attempt (#164) used negative ages on newborns but was intentionally reverted (#168) because it created phantom household members affecting SNAP and filing status.

Approach

Step 1: Stochastic assignment in cps.py

• In add_takeup(), use CDC WONDER age-specific birth rates to probabilistically assign is_pregnant = True to women aged 15-44
• Formula: P(pregnant | age) = (birth_rate / 1000) * (39/52) where 39/52 accounts for 9-month pregnancy duration within a year
• Use seeded_rng("is_pregnant") for reproducibility
• Store rates in parameters/take_up/pregnancy.yaml with state-level rates

Step 2: Calibration target via ETL

• Create etl_pregnancy.py following etl_medicaid.py pattern
• Target: person_count with is_pregnant == True constraint per state
• Source: CDC WONDER Natality birth counts by state
• Register "CDC WONDER Natality" as valid source in create_field_valid_values.py

Key design decisions

• Set current_pregnancies on the mother directly — do NOT create additional persons
• Birth counts are a reasonable proxy (miscarriages/abortions excluded but downstream programs mostly serve women who carry to term)
• State-level rates only (age-group cross would hit CDC suppression thresholds)

Closes #180, closes #193