feature: follow up to kernel reducers

[johnbowen42]

• Modify openMP backend to not perform copies of reducer tuples during reductions
• Re-assess if sequential params should be resolved inside the pattern/forall.hpp header
• Add tests combining reductions with a greater variety of kernel features, to assure compatiblity
• Benchmark hip/cuda/openmp backends
• Implement sycl backend for kernel param reductions