Reduction and multireduction refactoring

Brief summary:

• rewritten multireduction using lambda functions
• avoided volatile using __syncwarp()
• using reduction functions with return a + b instead of a += b
• using std::plus, std::multiplies, std::logical_and, std::logical_or, etc. instead of custom lambda functions
• optimized OpenMP thread counts for reduction and multireduction
• added computation of sample standard deviation to benchmarks
• implemented parallel prefix-sum with OpenMP
• implemented distributed prefix-sum