Try fix thread-safety

Author: XuehaiPan

include/pybind11/detail/internals.h

@@ -602,27 +602,26 @@ class internals_pp_manager {
     /// acquire the GIL. Will never return nullptr.
     std::unique_ptr<InternalsType> *get_pp() {
 #ifdef PYBIND11_HAS_SUBINTERPRETER_SUPPORT
-        if (get_num_interpreters_seen() > 1) {
-            // Whenever the interpreter changes on the current thread we need to invalidate the
-            // internals_pp so that it can be pulled from the interpreter's state dict.  That is
-            // slow, so we use the current PyThreadState to check if it is necessary.
-            auto *tstate = get_thread_state_unchecked();
-            if (!tstate || tstate->interp != last_istate_tls()) {
-                gil_scoped_acquire_simple gil;
-                if (!tstate) {
-                    tstate = get_thread_state_unchecked();
-                }
-                last_istate_tls() = tstate->interp;
-                internals_p_tls() = get_or_create_pp_in_state_dict();
+        // Whenever the interpreter changes on the current thread we need to invalidate the
+        // internals_pp so that it can be pulled from the interpreter's state dict.  That is
+        // slow, so we use the current PyThreadState to check if it is necessary.
+        auto *tstate = get_thread_state_unchecked();
+        if (!tstate || tstate->interp != last_istate_tls()) {
+            gil_scoped_acquire_simple gil;
+            if (!tstate) {
+                tstate = get_thread_state_unchecked();
             }
-            return internals_p_tls();
+            last_istate_tls() = tstate->interp;
+            internals_p_tls() = get_or_create_pp_in_state_dict();
         }
-#endif
+        return internals_p_tls();
+#else
         if (!internals_singleton_pp_) {
             gil_scoped_acquire_simple gil;
             internals_singleton_pp_ = get_or_create_pp_in_state_dict();
         }
         return internals_singleton_pp_;
+#endif
     }
 
     /// Drop all the references we're currently holding.