To activate the GPU path for force buffer ops, we should use two flags:
one indicating whether the feature is enabled (e.g. gpuBufferOpsEnabled) another indicating whether this step it should be used (e.g. useGpuBufferOpsWithReduction)
At least the latter should ideally be in the ppForceWorkload data structure which seems to be the natural place where we want to describe what work is happening on this rank.
- Also consider all possible combinations of triggers, and how to combine optimally in each case