Fix workgen tracking of table state across workloads

While chasing WT-4194, I ran into a bug where workgen was not tracking the number of records inserted into tables as expected. In particular, counts were reset to zero in each phase of the workload (each time a Workload object was created in Python).

I tracked this down to some code in ContextInternal::create_all that assumes sizeof (TableRuntime) == sizeof (uint64_t) – maybe that was true at some point, but not any more.