executors: fix CUDA constexpr dtor mismatch and memory ordering

- Restored non-constexpr dtor consistency for CUDA/NVCC builds.
- Upgraded mask reclamation to use acq_rel ordering for safe deletion.

Author: arpittkhandelwal

libs/core/executors/include/hpx/executors/parallel_executor.hpp

@@ -180,7 +180,7 @@ namespace hpx::execution {
                 first_core_ = rhs.first_core_;
                 num_cores_ = rhs.num_cores_;
                 auto* old_mask =
-                    mask_.exchange(nullptr, std::memory_order_relaxed);
+                    mask_.exchange(nullptr, std::memory_order_acq_rel);
                 delete old_mask;    // force recomputing cached pu mask
 
 #if defined(HPX_HAVE_THREAD_DESCRIPTION)
@@ -481,7 +481,7 @@ namespace hpx::execution {
             parallel_policy_executor&&) = default;
 
 #if defined(__NVCC__) || defined(__CUDACC__)
-        constexpr ~parallel_policy_executor() {}
+        ~parallel_policy_executor() {}
 #endif
 
     private: