Enhance OperationContext to ensure the transactional correctness of parallel scan stages

1. scan stages should see only the snapshot in which the original query was placed when it was started 
2. is it possible to resynchronize after yielding of single threads?
3. coordinate detaching of OperationContext from the main query to the parallel threads

Include writing unit tests stressing the locking part; e.g.

1. parallel scan with DOP > number_of_cores reading a collection that is being completely updated/inserted/removed by a separate write transaction
2. parallel scan with a concurrent DDL operation on the same collection (drop/create index/drop index)