verify-results-pgtap.sh can never accept a brand-new test file's first expected output

[jnasbyupgrade]

verify-results-pgtap.sh correctly scans for real not ok lines and plan mismatches first, but
then unconditionally fails make results if $TESTOUT/regression.diffs exists at all, regardless
of content:

# Also check regression.diffs (output mismatch even if pgtap all passed)
if [ -r "$TESTOUT/regression.diffs" ]; then
	echo "ERROR: Tests are failing. Cannot run 'make results'."
	...
	exit 1
fi

That makes it impossible to ever use make results to seed the first expected-output file for a
brand-new test: going from no baseline to any content is always a diff, even when every actual
pgtap assertion inside it says ok. The check can never pass for that case no matter how many
times it's rerun -- it's not a transient failure, it's structural.

Repro: add a new test/sql/foo.sql whose assertions all pass, run make test once so
test/results/foo.out exists, then make results. It refuses with "Tests are failing," even
though foo.out contains no not ok lines and no plan mismatch -- the only diff is
--- test/expected/foo.out not existing yet.

Worked around this time by hand-copying test/results/security.out to
test/expected/security.out directly, which the project's own docs otherwise say never to do.

Possible fix: when a given results/*.out file's expected/*.out counterpart doesn't exist yet,
skip the regression.diffs fallback for that file (the earlier not ok/plan-mismatch scan already
covers real regressions in its actual content).

[jnasbyupgrade]

Confirmed this is broader than the title suggests: the same block triggers for an existing file
that already had a working baseline, not just a brand-new one. Any accepted change to expected
output is by definition a diff between the old baseline and new actual output, so the unconditional
regression.diffs fallback check trips every time, regardless of whether every real TAP assertion
in the changed file passed. In practice this means make results can never succeed whenever there's
an actual change to accept -- only when there's nothing to update in the first place.

Repro this time: add a new pgtap assertion to an existing, already-passing test file, run make test once so the new line appears in test/results/*.out, then make results -- same "Tests are
failing" abort, on a file with zero not ok lines.