fix: listener retries with exponential backoff

Also corrects the admin ready response which now considers the listener
state.

test/io/test_io.py

@@ -733,8 +733,8 @@ def test_admin_ready_includes_schema_cache_state(defaultenv, metapostgrest):
 
         # force a reconnection so the new role setting is picked up
         postgrest.process.send_signal(signal.SIGUSR1)
-        # wait 600ms to finish schema cache reload attempt
-        time.sleep(0.6)
+
+        postgrest.wait_until_scache_starts_loading()
 
         response = postgrest.admin.get("/ready", timeout=1)
         assert response.status_code == 503

test/io/test_replica.py

@@ -20,7 +20,12 @@ def test_sanity_replica(replicaenv):
         response = postgrest.session.get("/items?select=count")
         assert response.text == '[{"count":10}]'
 
-    with run(env=replicaenv["replica"]) as postgrest:
+    working_replica_env = {
+        **replicaenv["replica"],
+        "PGRST_DB_CHANNEL_ENABLED": "false",  # LISTEN doesn't work on read replicas
+    }
+
+    with run(env=working_replica_env) as postgrest:
         response = postgrest.session.get("/rpc/is_replica")
         assert response.text == "true"