Rerun 95% GET latency test for 2.5.1 build, forcing quivalent behavior as STALE=UPDATE_AFTER

modify view query latency test to force STALE=UPDATE_AFTER behavior to allow comparison of performance 2.5.1 and 3.0.1-1440. use UPSERT before GET to force flush of buffer.

this issue originally written against kv latency test, but none of the kv tests set the parameter stale; and the 7 tests that do use the parameter stale=false, all reported in View Query -> Latency Query by Type do not exhibit the original concern, that 2.5.1 benefited from invalid implementation of "stale=false", as in each of the tests show improvement in latency 3.x over 2.5.1. now these are later 3.x tests than at the time of the original concern was logged, so likely the concern is no longer evident with the later 3.x builds.