US state carbon intensity still uses eGRID2016 data: update to eGRID2023 #1451

Problem

On private infrastructure in the US (no Electricity Maps token), CodeCarbon converts energy to emissions with codecarbon/data/private_infra/2016/usa_emissions.json, which holds eGRID2016 state CO₂ rates. EPA has since published eGRID2023 (rev. 2, June 2025), and the US grid has changed a lot:

45 of 51 states (incl. DC) are overstated, 40 of them by more than 10%. Three Northwest states (Idaho, Washington, Oregon) are understated by more than 10%.

Proposal

1. Replace the data with eGRID2023 state CO₂eq total output emission rates (STC2ERTA). CodeCarbon reports CO₂eq, and the old file used CO₂ only. The difference is under 2% for every state except Vermont, and I'm happy to use CO₂ (STCO2RTA) instead if you prefer.
2. Add a small script (scripts/update_usa_emissions.py) that builds the JSON from EPA's official workbook, so the next eGRID edition is a one-command update. This is a small step toward the "clear process for updating this data" in Dynamic WUE / PUE / carbon intensity data #947.
3. Update the methodology docs and the tests.

The file keeps the same schema (emissions in lbs/MWh + regionCode), so no code in emissions.py changes.

I'd like to work on this. I have a branch ready if this approach sounds good.