aggregation: use a string table to reduce field name storage when using aggregation

As an optimization, we can reduce the amount of allocated memory used by sharing the field name strings across all documents re-ified at the beginning of the aggregation pipeline. The string table can be put into the nascent ExpressionContext (we need to make sure that's available everywhere necessary). Every time we reify a document, use the strings in the string table, or add any missing ones.