Fix async Python functors invoking from multiple C++ threads (pybind#1587)

Ensure GIL is held during functor destruction.

Author: uentity

include/pybind11/functional.h

@@ -54,9 +54,20 @@ struct type_caster<std::function<Return(Args...)>> {
             }
         }
 
-        value = [func](Args... args) -> Return {
+        // ensure GIL is held during functor destruction
+        struct func_handle {
+            function f;
+            func_handle(function&& f_) : f(std::move(f_)) {}
+            func_handle(const func_handle&) = default;
+            ~func_handle() {
+                gil_scoped_acquire acq;
+                function kill_f(std::move(f));
+            }
+        };
+
+        value = [hfunc = func_handle(std::move(func))](Args... args) -> Return {
             gil_scoped_acquire acq;
-            object retval(func(std::forward<Args>(args)...));
+            object retval(hfunc.f(std::forward<Args>(args)...));
             /* Visual studio 2015 parser issue: need parentheses around this expression */
             return (retval.template cast<Return>());
         };