Backport reftest comparison machinery from our gtk4 packaging

This will let us distinguish between "fails by small differences caused
by rounding/i387" and "completely different result", without having to
move the whole build system to Meson, which seems like one variation too
many during a transition.

debian/log-reftests.py

@@ -0,0 +1,33 @@
+#!/usr/bin/python3
+# Copyright 2021 Simon McVittie
+# SPDX-License-Identifier: CC0-1.0
+
+import base64
+import sys
+from pathlib import Path
+
+if __name__ == '__main__':
+    for ui in Path('testsuite', 'reftests').glob('*.ui'):
+        for outputs in (
+            Path(
+                'debian', 'build', 'deb', 'testsuite', 'reftests',
+                'output',
+            ),
+        ):
+            diff = (outputs / (ui.stem + '.diff.png'))
+
+            if diff.exists():
+                ref = (outputs / (ui.stem + '.ref.png'))
+                out = (outputs / (ui.stem + '.out.png'))
+
+                for path in (ref, out, diff):
+                    if path.exists():
+                        print('')
+                        print('begin-base64 644 %s' % path)
+                        sys.stdout.flush()
+                        with open(path, 'rb') as reader:
+                            base64.encode(reader, sys.stdout.buffer)
+                        print('====')
+                        print('')
+
+                print('')