Fix: Thread sanitizer race.

See the comment in the source. This race should actually not be important, but we fix it anyways for now to make the CI happy.

lib/pls/include/pls/internal/data_structures/work_stealing_deque.h

@@ -21,10 +21,15 @@ using offset_t = stamped_integer::member_t;
 
 // Single Item in the deque
 class work_stealing_deque_item {
+  // TODO: In our opinion these atomic's are a pure formality to make the thread sanitizer happy,
+  //       as the race occurs in 'pop_head', where ALL CASES reading a corrupt/old value are cases
+  //       where the next CAS fails anywas, thus making these corrupted values have no influence on
+  //       the overall program execution.
+  //       ==> If we find performance problems in this queue, try removing the atoimcs again.
   // Pointer to the actual data
-  pointer_t data_;
+  std::atomic<pointer_t> data_;
   // Index (relative to stack base) to the next and previous element
-  offset_t next_item_;
+  std::atomic<offset_t> next_item_;
   offset_t previous_item_;
 
  public:
@@ -32,7 +37,7 @@ class work_stealing_deque_item {
 
   template<typename Item>
   Item *data() {
-    return reinterpret_cast<Item *>(data_);
+    return reinterpret_cast<Item *>(data_.load());
   }
 
   template<typename Item>
@@ -40,7 +45,7 @@ class work_stealing_deque_item {
     data_ = reinterpret_cast<pointer_t >(data);
   }
 
-  offset_t next_item() const { return next_item_; }
+  offset_t next_item() const { return next_item_.load(); }
   void set_next_item(offset_t next_item) { next_item_ = next_item; }
 
   offset_t previous_item() const { return previous_item_; }