c/controller_backend: try to force-abort reconfiguration only on leaders

Previously, when force-aborting a reconfiguration, we appended an
aborting configuration on all replicas. This can lead to log inconsistencies
as on followers the configuration will be duplicated (one from own append,
one replicated by the leader). Although these inconsistencies are
expected for force-abort, if the leader is alive, we can minimize the chance
of their appearance by waiting on followers for the aborting config to be
replicated from the leader.

Fixes redpanda-data#17847

(cherry picked from commit 8e221d3)

src/v/cluster/controller_backend.cc

@@ -1685,6 +1685,14 @@ controller_backend::force_abort_replica_set_update(
         }
         co_return errc::waiting_for_recovery;
     } else {
+        auto leader_id = partition->get_leader_id();
+        if (leader_id && leader_id != _self) {
+            // The leader is alive and we are a follower. Wait for the leader to
+            // replicate the aborting configuration, but don't append it
+            // ourselves to minimize the chance of log inconsistency.
+            co_return errc::not_leader;
+        }
+
         vlog(
           clusterlog.debug,
           "[{}] force-aborting reconfiguration",