fix: address code review feedback

- Fix 4 critical kernel bugs: use 3D blockId with uint64_t in simulator_cuda_kernels.h
- Create shared utilities in lib/util_cuda.h (GetBlockId, CreateGrid)
- Remove code duplication: eliminate duplicate CreateGrid functions
- Fix type safety in apps/qsimh_base_cuda.cu (std::stoul/std::stoull)
- Update all files to use shared 3D grid utilities

All changes verified with 32-qubit simulation test.

Author: sun-9545sunoj

apps/qsimh_base_cuda.cu

@@ -57,7 +57,7 @@ Options GetOptions(int argc, char* argv[]) {
   int k;
 
   auto to_int = [](const std::string& word) -> unsigned {
-    return std::atoi(word.c_str());
+    return std::stoul(word);
   };
 
   while ((k = getopt(argc, argv, "c:d:k:w:p:r:t:n:v:z")) != -1) {
@@ -66,28 +66,28 @@ Options GetOptions(int argc, char* argv[]) {
         opt.circuit_file = optarg;
         break;
       case 'd':
-        opt.maxtime = std::atoi(optarg);
+        opt.maxtime = std::stoul(optarg);
         break;
       case 'k':
         qsim::SplitString(optarg, ',', to_int, opt.part1);
         break;
       case 'w':
-        opt.prefix = std::atol(optarg);
+        opt.prefix = std::stoull(optarg);
         break;
       case 'p':
-        opt.num_prefix_gatexs = std::atoi(optarg);
+        opt.num_prefix_gatexs = std::stoul(optarg);
         break;
       case 'r':
-        opt.num_root_gatexs = std::atoi(optarg);
+        opt.num_root_gatexs = std::stoul(optarg);
         break;
       case 't':
-        opt.num_threads = std::atoi(optarg);
+        opt.num_threads = std::stoul(optarg);
         break;
       case 'n':
-        opt.num_dblocks = std::atoi(optarg);
+        opt.num_dblocks = std::stoul(optarg);
         break;
       case 'v':
-        opt.verbosity = std::atoi(optarg);
+        opt.verbosity = std::stoul(optarg);
         break;
       case 'z':
         opt.denormals_are_zeros = true;