Improvement idea: Make undistribute_table udf lazy

When we call `undistribute_table` [udf](https://github.com/citusdata/citus/pull/3913), it inserts whole the data from citus table to a new local table.

However, for `DISTRIBUTE_BY_NONE` citus tables (reference tables & citus local tables), we could introduce a lazy & fast logic, like we did when implementing `create_citus_local_table` udf.
`create_citus_local_table` creates the shard of the citus local table by simply suffixing `shardId` to the name of the given input table, and to its objects (indexes, constraints etc.).

We can follow a similar approach for `undistribute_table` udf as well. This time, we should de-suffix the `shardId` from the local shard relation name and also from the objects on it.

That way, we could execute `undistribute_table` udf for coordinator-replicated reference tables & citus local tables really quick, not depending on the relation size.