[FEA] The output columns of read_json need to follow depth-first-search order as in the input schema

[ttnghia]

Currently, read_json takes in the read schema as a map. As such, its column order cannot be preserved when the output table is generated. The callers need to keep track of the column order by themselves and rearrange the columns based on the output column names.

We should support preserving the column order (by depth-first-search order) so the callers can reduce their overhead of doing so. In order to do so, the input schema needs to be specified using a std::vector of nested columns instead of std::map.