Fix Windows build: use idx_t for OpenMP loop variable

  MSVC requires signed integral type for OpenMP loop variables.
  Changed from size_t to idx_t in compute_residuals().

Author: Ray0907

faiss/IndexIVFPQ.cpp

@@ -152,7 +152,7 @@ static std::unique_ptr<float[]> compute_residuals(
     std::unique_ptr<float[]> residuals(new float[n * d]);
     // Parallelize with OpenMP (each iteration is independent)
 #pragma omp parallel for if (n > 1000)
-    for (size_t i = 0; i < n; i++) {
+    for (idx_t i = 0; i < n; i++) {
         if (list_nos[i] < 0)
             memset(residuals.get() + i * d, 0, sizeof(float) * d);
         else