Count fieldname bytes against the Postgres identifier limit

Long non-ASCII custom field names no longer get truncated on Postgres, which caused duplicate-column failures on every later schema sync.