There are many places in this test program that need to consume a
PGresult while checking that its PQresultStatus is as-expected, or
related tasks such as checking that PQgetResult has nothing more to
return.  These tasks were open-coded in a rather inconsistent way,
leading to some outright bugs, some memory leakage, and frequent
inconsistencies about what would be reported in event of an error.
Invent a few helper functions to standardize the behavior and
reduce code duplication.  Also, rename the one pre-existing helper
function from confirm_query_canceled to consume_query_cancel, per
Álvaro's suggestion that "confirm" is a poor choice of verb for a
function that will discard the PGresult.

While at it, clean up assorted other places that were leaking
PGresults or even server connections.  This is pure neatnik-ism,
since the test doesn't run long enough for those leaks to be of
any real-world concern.

While this fixes some things that are clearly bugs, it's only
a test program, and none of the bugs seem serious enough to
justify back-patching.

Bug: #18960
Reported-by: Dmitry Kovalenko <d.kovalenko@postgrespro.ru>
Author: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: Álvaro Herrera <alvherre@kurilemu.de>
Discussion: https://postgr.es/m/18960-09cd4a5100152e58@postgresql.org