c/controller_backend: allow shutdown_partition to fail on app shutdown

With `ss::gate_closed_exception` only. This is to untangle a race
condition between partition shutdown and app shutdown. Callers should
consider the operation unsuccessful and will probably retry it, as
`shard_placement_table` is not going to be updated.

While this unsuccessful operation will leave the partition in a somewhat
orphaned state, i.e. without appropriate `shard_table` entries, the rest
of the system should be able to live with it in the same way as it lives
during a successful partition shutdown.

(cherry picked from commit 0bb1be6)

src/v/cluster/controller_backend.cc

@@ -1924,10 +1924,17 @@ controller_backend::shutdown_partition(ss::lw_shared_ptr<partition> partition) {
           ntp, gr, partition->get_log_revision_id());
         // shutdown partition
         co_return co_await _partition_manager.local().shutdown(ntp);
+    } catch (const seastar::gate_closed_exception&) {
+        // let it bubble up and break reconciliation loop
+        vlog(
+          clusterlog.debug,
+          "error shutting down {} partition, app shutting down",
+          ntp);
+        throw;
     } catch (...) {
         /**
-         * If partition shutdown failed we should crash, this error is
-         * unrecoverable
+         * If partition shutdown failed with any other exception we should
+         * crash, this error is unrecoverable
          */
         vassert(
           false,

src/v/cluster/controller_backend.h

@@ -302,6 +302,7 @@ class controller_backend
     ss::future<> remove_from_shard_table(
       model::ntp, raft::group_id, model::revision_id log_revision);
 
+    /* may fail only with ss::gate_closed_exception */
     ss::future<xshard_transfer_state>
       shutdown_partition(ss::lw_shared_ptr<partition>);