fix: apply the pg keepalive settings to the postgres trigger connection (#11512)

Claude-Session: https://claude.ai/code/session_019cpWeGy8jRMtWqvuRBpgu4

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

backend/windmill-common/src/lib.rs

@@ -1072,7 +1072,7 @@ impl TokioPgConnection {
 /// Without these, a server that vanishes without closing the socket (a failover,
 /// a dropped route) leaves a query waiting on a read for the OS default of two
 /// hours. The server's kernel answers the probes, so a slow query is unaffected.
-fn set_pg_keepalive(config: &mut tokio_postgres::Config) {
+pub fn set_pg_keepalive(config: &mut tokio_postgres::Config) {
     config
         .keepalives(true)
         .keepalives_idle(std::time::Duration::from_secs(60))

backend/windmill-trigger-postgres/src/lib.rs

@@ -355,6 +355,8 @@ pub async fn get_raw_postgres_connection(
         config.options(options);
     }
 
+    windmill_common::set_pg_keepalive(&mut config);
+
     let connector = build_tls_connector(ssl_mode, database.root_certificate_pem.as_ref())?;
     let client = if let Some(connector) = connector {
         let (client, connection) = config.connect(connector).await.map_err(to_anyhow)?;