Skip old queries on search shards

Changes the `search_shards` api to skip sending unsupported queries,
instead sending `match_all` which should make it "fail open", returning
"this shard can match this query" for all unsupported queries.

This is *weird*, but it's useful for systems that build the query
internally inside of elasticsearch and can fall back to filtering
outside of lucene. Which is *exactly* what ESQL does.

Author: nik9000

server/src/main/java/org/elasticsearch/action/search/SearchShardsRequest.java

@@ -16,6 +16,7 @@
 import org.elasticsearch.common.io.stream.StreamInput;
 import org.elasticsearch.common.io.stream.StreamOutput;
 import org.elasticsearch.core.Nullable;
+import org.elasticsearch.index.query.MatchAllQueryBuilder;
 import org.elasticsearch.index.query.QueryBuilder;
 import org.elasticsearch.tasks.Task;
 import org.elasticsearch.tasks.TaskId;
@@ -77,7 +78,16 @@ public void writeTo(StreamOutput out) throws IOException {
         super.writeTo(out);
         out.writeStringArray(indices);
         indicesOptions.writeIndicesOptions(out);
-        out.writeOptionalNamedWriteable(query);
+        out.writeOptionalNamedWriteable(
+            query == null || query.supportsVersion(out.getTransportVersion()) ? query :
+            /*
+             * The remote node doesn't support the query we're sending. If this were only
+             * used for _search this could just fail, but for ESQL it's much more convenient
+             * if it pretends that the query is MatchAll. ESQL will frequently be able to
+             * perform the document filtering on the data node in its engine. Slowly.
+             * But correctly.
+             */
+            new MatchAllQueryBuilder());
         out.writeOptionalString(routing);
         out.writeOptionalString(preference);
         out.writeBoolean(allowPartialSearchResults);