Stop supporting Canada, Nigeria, and Israel country models

Proposal

Drop ca, ng, and il from the household API's supported countries, leaving us and uk.

Motivation

• Every worker loads all five country models at startup; policyengine-canada, policyengine-ng, and policyengine-il add image size, memory, snapshot build time, and dependency-resolution surface for country routes with no known partner traffic.
• The packages are not release-significant (only US and UK versions participate in Modal release naming, manifests, and validation per release_config.py), so their pins drift untested — the same silent-staleness failure mode that let the UK path break unnoticed.
• Removing them simplifies the planned UK/US app split (see the UK-extrication issue): the "standard" app becomes a US app outright.

What removal touches

• COUNTRIES in libs/household-api/policyengine_household_api/country.py and COUNTRY_PACKAGE_VERSIONS / COUNTRIES in constants.py, plus the current_law_id map in build_metadata.
• Dependency pins in libs/household-api/pyproject.toml (policyengine-canada, policyengine-ng, policyengine-il) and uv.lock.
• validate_country and the OpenAPI spec's country enumeration; docs mentioning supported countries.
• Decide the HTTP contract for retired routes: 404 from validate_country (natural once removed) vs. a 410 Gone with a message for any existing caller.

Before removing, check analytics for any real traffic to /ca/*, /ng/*, or /il/* calculate routes.

Start by checking analytics for traffic to /ca/, /ng/, and /il/* calculate routes, then read country.py, constants.py, build_metadata, and release_config.py. Review the dependency pins in pyproject.toml and uv.lock, validate_country, the OpenAPI country enumeration, and supported-country docs. Done means the three countries are removed consistently and the retired-route response contract is decided and applied.