Extend schema to contain table name, and migrate cost model's ColumnRef property usage to schema

[jurplel]

From Chi:

yes we should do that, feel free to create an issue...
this is so-called "column name inference"
assign a name for unnamed columns, and use user-specified names for projections
and for the top-level node, if we have physical property implemented, it will always be a gather node
unless all plan nodes are singleton

ColumnRef and Schema seem to have a lot of purpose overlap, I think that we can merge their functionality together.

[jurplel]

After reading more into ColumnRef, I am not as sure that it is redundant... Seems like a lot going on

[skyzh]

yes, they are all column-level properties; but they are different -- schema is more focused on column type + name, while column ref indicates how the column is computed (which will then be used by the cost model to derive the cost)

[skyzh]

with #202, I feel the final state of this property will be like:

• after aggregation: we will get something like Distinct(ColumnRef)
• after projection, i.e., select v1 + 1 from table: we will get something like Add(v1, 1)
• after join, we will get something like Correlated(X, Y, original_column_ref)

so it could be more complex that what it is right now