chore(naming): align Layer-1 target fixtures to lr_ged_* (Layer-2 pred_ln_* excluded)

[Polichinel]

Companion to views-platform/views-models#151 (standardize the regression-target name to lr_ged_{sb,ns,os}).

Scope

• Update test fixtures referencing lr_*_best (~10 occurrences) to the canonical lr_ged_* where they denote the Layer-1 model target.

Explicitly OUT of scope

views_postprocessing/unfao/managers/unfao.py:223 hardcodes the forecast delivery targets ["pred_ln_sb_best", "pred_ln_ns_best", "pred_ln_os_best", "pred_ln_sb_prob", "pred_ln_ns_prob", "pred_ln_os_prob"]. That is a Layer-2 delivery schema, not the Layer-1 model-target name — leave it for the separate FAO no-collapse / delivery-contract work.

Canonical: lr_ged_sb / lr_ged_ns / lr_ged_os.

[Polichinel]

Done on development (PR #50, commit 3ff5d95) — closing.

Note on scope: the lr_*_best pattern this issue described did not exist in this repo (verified across all tracked files including the .npz fixture keys). The real Layer-1 target straggler was ged_sb_best (3× in tests/test_views_frames_conformance.py) — which both views-models#151 and this repo's docs/ADR-011_implementation_assessment.md map to the canonical lr_ged_sb. Renamed those 3; left the Layer-2 pred_ln_sb_best delivery column untouched (out of scope). ruff clean, full suite green.

(GitHub didn't auto-close on the PR because it merged to development, not the default branch main — closing manually. It'll reach main with the next release.)