Cf. https://www.postgresql.org/message-id/20210506162651.GJ27406@telsasoft.com

Newer distributions work without it, just add it unconditionally.

This appears to provoke stack overflows when testing MonetDB.

Cleaning it is a bad idea when building from a release tarball.

Reported-by: @alexanderlaw
Closes #19

filter.sql was replaced by known.txt and known_re.txt.

See https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=fd1a421fe66173fb9b85d3fe150afde8e812cbe4

Blind attempt at increasing generation speed.

Per input from @mlkersten.

This reduces the MonetDB error rate from 90% to 50%.  MonetDB doesn't
appear to support LIMIT clauses in atomic subselects.

The previous commit surprisingly yielded a CPU hogging sqlsmith with
monetdb due to the table_or_query_name production hardly finding
candidates.  It turned out that the scope model was seriously lacking:
Creating a child scope now automatically includes the table and column
references available to the parent scope.  Previously, child scopes
only contained table and column references explicitly put there by the
parent production.