Un-break jsonb support in django newer than 3.1.1

Django 3.1.1 intentionally broke jsonb support for all queries not going
through the ORM. And of course we have those, since trying to write some
of these things in the ORM is if not impossible then at least a recipie
for insanity and unreadable code.

Fix is to specifically turn the jsonb support back on for the queries we
run outside the ORM. This needs to doubly-unwrap debug cursors if we're
in debug mode, but at least it appears to work on both 2.2 and 3.2 in
the same way.

web/pglister/util.py

 from django.db import connection
 
+import psycopg2.extras
 
-def exec_to_list(query, params=None):
+
+def _get_cursor():
     curs = connection.cursor()
+
+    # Django debug mode will wrap the cursor. And then the django debug toolbar will wrap it
+    # again. So unwrap until we have an actual psycopg2 cursor.
+    psycocurs = curs
+    while hasattr(psycocurs, 'cursor'):
+        psycocurs = psycocurs.cursor
+
+    # Django 3.1.1 (YES! In a security update!!) intentionally breaks PostgreSQL jsonb support
+    # for access that's not through the ORM. Un-break this by turning the support back on,
+    # but do so specifically for the cursor that we're using in this query only, so that django
+    # can keep the broken behavior for its cursors.
+    psycopg2.extras.register_default_jsonb(conn_or_curs=psycocurs)
+    return curs
+
+
+def exec_to_list(query, params=None):
+    curs = _get_cursor()
     curs.execute(query, params)
     return curs.fetchall()
 
 
 def exec_to_row(query, params=None):
-    curs = connection.cursor()
+    curs = _get_cursor()
     curs.execute(query, params)
     return curs.fetchone()
 
 
 def exec_to_dict(query, params=None):
-    curs = connection.cursor()
+    curs = _get_cursor()
     curs.execute(query, params)
     columns = [col[0] for col in curs.description]
     return [dict(zip(columns, row))for row in curs.fetchall()]
 
 
 def exec_to_scalar(query, params=None):
-    curs = connection.cursor()
-    curs.execute(query, params)
+    curs = _get_cursor()
     r = curs.fetchone()
     if r:
         return r[0]
@@ -38,7 +56,7 @@ def exec_to_hier(query, params=None):
     # XXX: make configurable to do this at more than one level?
     # XXX: assumes that query comes sorted by first columns, and that
     #      the first column is unique.
-    curs = connection.cursor()
+    curs = _get_cursor()
     curs.execute(query, params)
     columns = [col[0] for col in curs.description[1:]]
     data = {}