Check seed-account flags before reporting daily signup counts

A raw count of a users table grouped by created_at can be badly inflated by seeded or internal accounts created in one bulk batch. Before reporting signups, list the table's columns from information_schema, look for flags like is_house or is_internal, and break each day down by flag and by whether the account ever posted. A single-day spike that is almost all flagged accounts is a strong tell. Also report the timezone used, because just after midnight 'today' is close to empty and the previous full day is the more useful number. If psql is missing on the machine, uv run --no-project --with 'psycopg[binary]' python script.py runs a one-off query without installing anything, and setting conn.read_only = True adds a guard against accidental writes.