Spread CPS age topcode across 80 to 84

[MaxGhenis]

Summary

Fixes CPS ASEC age topcoding for MP's eCPS-shaped path. Census A_AGE == 80 represents ages 80-84, but MP had been exporting every such person as exactly age 80. The latest MP-vs-eCPS target diagnostics showed this as the top individual regression: MP estimated about 10.27M weighted people at exact age 80 versus a target around 1.77M, while eCPS spreads that topcoded group across ages 80-84.

This PR:

• adds an MP-owned deterministic CPS age recode helper, without importing policyengine-us-data at runtime;
• applies it in the main CPS source processing path before age-derived flags are created;
• applies it in the older CPS mapping/transform helpers as well;
• bumps the processed CPS cache version so local rebuilds do not reuse stale age=80 caches;
• adds focused tests for deterministic 80-84 spreading and raw CPS load behavior.

Validation

• uv run ruff check src/microplex_us/data_sources/cps.py src/microplex_us/data_sources/cps_age.py src/microplex_us/data_sources/cps_mappings.py src/microplex_us/data_sources/cps_transform.py tests/data_sources/test_cps_age.py
• uv run ruff format --check src/microplex_us/data_sources/cps.py src/microplex_us/data_sources/cps_age.py src/microplex_us/data_sources/cps_mappings.py src/microplex_us/data_sources/cps_transform.py tests/data_sources/test_cps_age.py
• uv run --extra dev python -m pytest tests/data_sources/test_cps_age.py -q
• uv run --extra dev python -m pytest tests/test_cps_source_provider.py -q -k 'rebuilds_stale_processed_cache_without_state_fips or caches_household_geography_on_persons'