core: DAG, revisited

[skyzh]

though we know the core can process DAG queries now, we don't know how efficient it is. for example, if two parts of the query could reuse one component, currently, it's likely that we will produce the plan with that query component appearing twice. the bridge should insert materialize node, and the core should have a way to let the bridge know where to insert the node, and let the bridge make the decision (?)

also, the cost model needs to be improved. currently, we will count the cost of the materialize node twice, because in the memo table, it's considered as two entities in the final plan.