In JDK-8298597, we deferred the work of making merge() work with decaying stats. This requires two things:

1- a slight rethinking of the API (or its extension) to deal correctly with the concept of "rounds of temporally synchronous samples"
2- fitting the above into Tony Finch's incremental online computation of decaying statistics
3- adjusting the associated gtest (see aforementioned ticket)

This was not needed for the only use case to date (in GenShen), so was deferred for the sake of expediency.