domrf_kn_objects retains ~9 snapshots per obj_id (prod: 4090 rows / 482
distinct obj_id for Строящиеся = 8.49× inflation). _active_competitors_count
and developer_portfolio counted every historical snapshot as a distinct
object; _competitors_two_dim's active CTE had the same class. Dedup via
DISTINCT ON (obj_id) ORDER BY snapshot_date DESC, applying VOLATILE filters
(site_status, district_name, obj_class — prod variance 11/1/29) in the outer
query after dedup so an object is counted by its TRUE-latest snapshot, while
stable scope (region_cd, dev_id — variance 0) stays in the CTE. Competitor
counts and developer portfolios drop ~8× — correct de-inflation.
Also backfill ON CONFLICT DO UPDATE SET: UPSERT_OBJECT_SQL dropped 8
freshly-computed columns on re-scrape, UPSERT_PHOTO_SQL dropped build_type —
added as direct EXCLUDED (not COALESCE, so problem_flag can clear to NULL).
Refs #2464 (clusters E + F).
