k/metadata: do not return no leader when leader is not yet available

Instead of returning `-1` indicating that there is an ongoing leader
election we return randomly selected node as a partition leader. This is
much less interrupting for the client as it simply forces it to refresh
metadata. Some clients do not tolerate `-1` returned from Metadata
handler and simply stop working.

Signed-off-by: Michal Maslanka <michal@redpanda.com>

src/v/kafka/server/handlers/metadata.cc

@@ -80,8 +80,14 @@ std::optional<cluster::leader_term> get_leader_term(
   const cluster::metadata_cache& md_cache,
   const std::vector<model::node_id>& replicas) {
     auto leader_term = md_cache.get_leader_term(tp_ns, p_id);
+    /**
+     * If current broker do not yet have any information about leadership we
+     * simply return random node to force the client metadata update.
+     */
     if (!leader_term) {
-        return std::nullopt;
+        auto idx = fast_prng_source() % replicas.size();
+        leader_term.emplace(replicas[idx], model::term_id(-1));
+        return leader_term;
     }
     if (!leader_term->leader.has_value()) {
         const auto previous = md_cache.get_previous_leader_id(tp_ns, p_id);