Column-grammar migration: adopt ShapePipe-v2 column names (ngmix_v2.0) #218

Sub-issue of the ngmix v2.0 epic (shapepipe#762), under shapepipe#777 (the column-grammar PR, #761).

shapepipe#761 renames every shape-measurement column onto one grammar (ESTIMATOR_COMPONENT_OBJECT[_metacaltype]), stores a single size T = 2σ² via cs_util.size, drops GAL/spread_model, and restores a genuine original-PSF fit (*_PSF_ORIG, distinct from the reconvolved-kernel *_PSF_RECONV, per shapepipe#749). This issue tracks propagating that into sp_validation:

• Rename every consumer site to the shipped grammar (NGMIX_G1/G2_{shear}, NGMIX_T_PSF_RECONV_{shear}, NGMIX_G1/G2/T_PSF_ORIG_{shear}, HSM_G1/G2/T/FLAG_{PSF,STAR}) — galaxy.py, catalog.py, calibration.py, glass_mock.py, scripts/calibration/extract_info.py, cosmo_val/cat_config.yaml, config/calibration/mask_v1.X.*.yaml.
• Route every size conversion through cs_util.size; drop the manual SIGMA**2 / square_size handling now that stored sizes are already the area T.
• Remove the spread_model-based classification path (shapepipe no longer writes those columns).
• Re-validate (not assume) the *_PSF_ORIG value change against a catalogue regenerated after #741 merges.
• Refresh docs/ngmix_psf_column_migration.md, which predates the final "drop GAL" decision and is currently wrong.

Implemented by: draft PR #201 (branch migrate/ngmix-psf-column-names).

The e/g ellipticity-convention question raised during shapepipe#761's review does not require any action here — it resolved to "both estimators already store native g," so no conversion step is needed downstream.

— Claude on behalf of Cail