Fix numbering of PostgreSQL params across reusable queries

[Niols]

When a reusable query (-- @foo | include: reuse) is spliced into an outer query as a CTE via WITH x AS &ref, the reused query's parameters and the outer query's parameters end up in a single generated statement, bound positionally by set_params.

With -params postgresql, the positional counter was reset to 0 for the spliced fragment and left unchanged when the outer query resumed, so both fragments numbered their $N placeholders starting from $1. The outer parameters therefore collided with the reused ones. Example:

-- @get_persons | include: reuse
SELECT * FROM person WHERE name LIKE @name;

-- @get_users 
WITH person AS &get_persons
SELECT * FROM "user" JOIN person ON person.user_id = "user".id
WHERE "user".id > @min_id;

On such an input, -params postgresql generated:

WITH person AS (... WHERE name LIKE $1)
SELECT ... WHERE "user".id > $1

Note the $1 used twice.

(This was actually caught by PostgreSQL because set_params still supplies a value for every position, so I got “bind message supplies N parameters, but prepared statement requires M”.)

I checked this only affects the postgresql parameter style; others use names (named and oracle) or use positional ? (unnamed).

A first commit adds tests exhibiting this behaviour; a second threads the positional counter through the shared-query splice in substitute_vars.