race in Primproc-side hash table construction code

Greg found this one doing some benchmarks on large-ish tables.

Occasionally a query involving a join will hang forever.

Looked into it, found it was a race on a counter used to indicate that PrimProc has inserted all of the rows for the small-side tables. Only affects string/compound joins; that is the only kind that modify the counter, and would only happen when there are enough rows s.t. many msgs with the join data are issued and processed at once.

Made the counter an atomic var to eliminate the race.