fix(event-loop): timers with a complicated job function can't be automatically unref'ed sometimes

See issue: heap-use-after-free in timerJobWrapper (#341)

We need to ensure the memory block doesn't move for reallocation before we can use the pointer to the timer's `AsyncHandle`, as we could have multiple new `setTimeout` calls to expand the `_timeoutIdMap` vector while running the job function in parallel.

Author: Xmader

src/PyEventLoop.cc

@@ -38,13 +38,13 @@ static PyObject *timerJobWrapper(PyObject *jobFn, PyObject *args) {
   PyEventLoop::AsyncHandle::id_t handleId = PyLong_AsLong(PyTuple_GetItem(args, 1));
   double delaySeconds = PyFloat_AsDouble(PyTuple_GetItem(args, 2));
   bool repeat = (bool)PyLong_AsLong(PyTuple_GetItem(args, 3));
-  auto handle = PyEventLoop::AsyncHandle::fromId(handleId);
 
   PyObject *ret = PyObject_CallObject(jobFn, NULL); // jobFn()
   Py_XDECREF(ret); // don't care about its return value
 
   PyObject *errType, *errValue, *traceback; // we can't call any Python code unless the error indicator is clear
   PyErr_Fetch(&errType, &errValue, &traceback);
+  auto handle = PyEventLoop::AsyncHandle::fromId(handleId); // Making sure a `AsyncHandle::fromId` call is close to its `handle`'s use. We need to ensure the memory block doesn't move for reallocation before we can use the pointer, as we could have multiple new `setTimeout` calls to expand the `_timeoutIdMap` vector while running the job function in parallel.
   if (repeat && !handle->cancelled()) {
     _enqueueWithDelay(_loop, handleId, jobFn, delaySeconds, repeat);
   } else {