Fix dx4 calculation in scalar remainder code and add null checks

- Fix dx4 computation in scalar remainder loops (should be dx2*dx2)
- Add missing null pointer check in benchmark_sinf for consistency

Co-authored-by: wvbbreu <185333235+wvbbreu@users.noreply.github.com>

benchmarks/benchmark_utils.hpp

@@ -26,6 +26,9 @@ static void benchmark_sinf(benchmark::State &state) {
       reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
   float *s =
       reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
+  if (!x || !s) {
+    throw std::runtime_error("Buffer allocation failed");
+  }
   auto end = std::chrono::high_resolution_clock::now();
   state.counters["init_ms"] =
       std::chrono::duration_cast<std::chrono::microseconds>(end - start)