Reducer Tuning Leftovers

[MrBurmark]

There are a few things leftover from the reduction tunings PR #1625.

• Add support in the new reducer interface for tuning options, this interface should allow single reduction trees and coalesced atomics when multiple reducers are used
• Consider using for_each_type to expand type lists in a test instead of creating multiple tests to improve compile times, for example over reduce policies
• Consider adding a config flag so the default atomic policy can be either a host init policy or not out of concern for systems without a good way to access device memory from the host
• Consider adding a config flag so the hip intrinsic code can be compiled out of an abundance of caution