Parallel Query: Open tables for worker threads

Worker threads will need their own handler (i.e. ha_innobase, ha_maria, etc) objects.

This is because each worker should have its own "cursor" (i.e. current position in the scan) and table->record[0] (the current record). The objects should share the same read view of the database.

Should TABLE object itself be shared?

Should TABLE_LIST object be shared? (probably not as it's a parse tree-level entity)

table->field[0...n] should not be shared, as those have pointers to table->record[0] which
should not be shared.

How to create additional handler objects

Option 1: open table normally

Open tables "normally" by calling open_table() (or what is the function to get TABLE object
from open table cache). We will get new TABLE and handler objects.

Option 2: handler::clone

Use handler->clone() to get another handler.

It is not clear if/how we would get another TABLE object.