Defeat shared_ptr's single-threaded optimizations

WeaselDB is always going to start multiple threads, so we don't care
about single-threaded performance

benchmarks/bench_reference.cpp

@@ -43,9 +43,16 @@ template <typename T> struct PointerTraits<Ref<T>> {
   static const char *weak_name() { return "WeakRef"; }
 };
 
+// Force multi-threaded mode to defeat __libc_single_threaded optimization
+void force_multithreaded() {
+  std::thread t([]() {});
+  t.join();
+}
+
 template <typename PtrType>
 void benchmark_creation(ankerl::nanobench::Bench &bench) {
   using Traits = PointerTraits<PtrType>;
+  force_multithreaded();
 
   bench.run(std::string(Traits::name()) + " creation", [&] {
     auto ptr = Traits::make(TestObject{123});
@@ -56,6 +63,7 @@ void benchmark_creation(ankerl::nanobench::Bench &bench) {
 template <typename PtrType>
 void benchmark_copy(ankerl::nanobench::Bench &bench) {
   using Traits = PointerTraits<PtrType>;
+  force_multithreaded();
 
   auto original = Traits::make(TestObject{123});
   bench.run(std::string(Traits::name()) + " copy", [&] {
@@ -80,6 +88,7 @@ void benchmark_move(ankerl::nanobench::Bench &bench) {
 template <typename PtrType>
 void benchmark_weak_copy(ankerl::nanobench::Bench &bench) {
   using Traits = PointerTraits<PtrType>;
+  force_multithreaded();
 
   auto strong_ptr = Traits::make(TestObject{123});
   typename Traits::weak_type weak_original = strong_ptr;