Introduce use_cornerstone_ghg_model bucket flag (GHG model allocation)

[briantobin-99]

cc:
Closes:

What changed? Why?

Introduces the "GHG model allocation" bucket flag and dissolves the year-locked inventory flags into the data-year knob:

• New use_cornerstone_ghg_model: ON = load the pre-built GHG_national_Cornerstone_{usa_ghg_data_year} FBS parquet from GCS (with the eGRID variant when implement_electricity_disaggregation is on); OFF = the GHG_national_CEDA_{year} flowsa FBS (legacy CEDA allocation methodology; raises a clear error for years ≠ 2023, where no CEDA method file exists).
• Absorbs new_ghg_method (identical mechanism) and v0_3_umd_2024_ghgia (the same mechanism year-locked to 2024). Which inventory and attribution vintages a year carries (EPA GHGI vs UMD GHGIA, MECS survey year) is defined by the per-year method files in transform/ghg/, not by config flags.
• Deletes update_mecs_method and v0_3_umd_2023_ghgia — intermediate 2023 inventory states that never shipped — along with their now-unselectable variant method files (GHG_national_Cornerstone_2023_{ghgi_mecs,umd_ghgia}.yaml) and the duplicated year-guard blocks (_select_flowsa_ghg_method is gone; selection lives in one place in load_E_from_flowsa).
• The mutual-exclusion validator with use_ghg_national_2023_m2 carries over to the new flag.

Config migration: 2025_usa_cornerstone_v0_3.yaml, the v03_waterfall_* configs, and the electricity configs swap their new_ghg_method / v0_3_umd_2024_ghgia keys for use_cornerstone_ghg_model. Under the new scheme, the waterfall G2→G3 diff ("US data update") is purely usa_ghg_data_year: 2023 → 2024.

Testing

Unit suite, black, ruff, and mypy green. derive_E_usa() under 2025_usa_cornerstone_v0_3 matches the pinned E_USA_ES snapshot exactly (max abs diff 0.0). Snapshot integration suite (test_integration run) passes on the stack including this change.

[bl-young]

Should this be true by default?

[bl-young]

if we do set the default to be True this will need to be revised