Thanks to github user nikhil-postgres for the report.

While at it also fix a thinko in the size estimation which lead to greatly
overestimated hypothetical hash index sizes.