* Allow missing RUN in CLUSTER functions, which would mean `RUN ON 0;`
This should be useful for single-partition clusters.
+ * Lazy value type cache, per-conn binary I/O decision.
+
== Just thoughts ==
* A way to give config to CONNECT functions.
* Streaming for big resultsets, to avoid loading them fully in memory.
This needs also backend changes.
* Drop CONNECT, SELECT, plproxy.get_cluster_config()...
+ * Debug a way to survive AS syntax with different types.