application: stop schema_registry clients before datalake subsystems

We've seen a shutdown hang that appears to be caused by a datalake
translator getting stuck repeatedly trying to query the schema registry.

This attempts to fix this by moving schema registry client shutdown to
before datalake shutdown.

Author: andrwng

src/v/pandaproxy/schema_registry/api.cc

@@ -112,6 +112,10 @@ ss::future<> api::start() {
     co_await _service.invoke_on_all(&service::start);
 }
 
+ss::future<> api::stop_clients() {
+    co_await _client.invoke_on_all(&kafka::client::client::stop);
+}
+
 ss::future<> api::stop() {
     vlog(srlog.debug, "Stopping schema registry API...");
     co_await _client.invoke_on_all(&kafka::client::client::stop);

src/v/pandaproxy/schema_registry/api.h

@@ -49,6 +49,7 @@ class api {
 
     ss::future<> start();
     ss::future<> stop();
+    ss::future<> stop_clients();
     ss::future<> restart();
 
     const configuration& get_config() const;

src/v/redpanda/application.cc

@@ -374,6 +374,11 @@ void application::shutdown() {
             return cloud_io.invoke_on_all(&cloud_io::remote::request_stop);
         });
     }
+    // Shutdown schema registry clients before the datalake subsystems to
+    // ensure the datalake subsystems can shutdown quickly.
+    if (_schema_registry) {
+        _schema_registry->stop_clients().get();
+    }
     /**
      * Shutdown the datalake services before stopping all the partitions.
      * NOTE: translators may call into the coordinator via the coordinator