Refactor the TAP test of auto_explain

Refactor the TAP test of auto_explain

Previously, the tests were structured so as all the queries whose plans
are checked run first, followed by pattern checks using the full set of
server logs.  This can be problematic when extending the tests, as this
increases query plan overlaps, where two tests finish by having similar
plan outputs potentially invalidating the tests wanted.

The tests are refactored so as log content matches are checked in
isolation of each query run, by grabbing the position of the server logs
before running each query whose plan is generated in the logs.  This
avoids issues when extending the tests, something that would become a
larger problem with a follow-up patch that adds a new GUC in
auto_explain to control the size of the each parameter logged.

Author: Dagfinn Ilmari Mannsåker
Discussion: https://postgr.es/m/[email protected]