Memory issues when adding a new function with pybind11

In attempting to add a new function, callable from Python, similar to electron_count, the function gives a segmentation fault on returning an ElectronCountedData object if no py::array_t objects are included as parameters to that function. However, if one or more py::array_t arrays are given as parameters, it seems to work without problems.

The addition of the function follows the general methodology used in: https://github.com/OpenChemistry/stempy/blob/master/python/image.cpp

A simple example is given here: https://github.com/jerenner/stempy/tree/example
(diff is here: https://github.com/jerenner/stempy/commit/60ae39090287c3f083f120e6b0df16c6fdac4cec)

The example runs the following code:

{
  Dimensions2D scanDimensions = { 0, 0 };
  Dimensions2D frameSize = { 0, 0 };

  Events events;
  events.resize(10);

  ElectronCountedData ret;
  ret.data = events;
  ret.scanDimensions = scanDimensions;
  ret.frameDimensions = frameSize;

  return ret;
}

using 2 different methods, one which takes a py::array_t and the other that does not.

1. ElectronCountedData testMethodBasic(int test)
2. ElectronCountedData testMethodArray(int test, const float test_arr[])

A Jupyter notebook showing the functions run in Python is here: https://github.com/jerenner/4dstem/blob/main/pybind_test.ipynb

For some reason (which could be machine-dependent, as this has only been tested on one machine), the method that takes the array as a parameter does not produce any problems, but the one without any array parameters gives a segmentation fault, which appears to occur on line 28 of https://github.com/OpenChemistry/stempy/blob/master/python/image.cpp:

return py::array(ptr->size(), ptr->data(), capsule);

Start with python/image.cpp around the reported line 28 and reproduce the crash using the example repository, commit 60ae390, and pybind_test.ipynb. Compare testMethodBasic with testMethodArray and trace the returned ElectronCountedData and array handling. Done means both functions run from Python without a segmentation fault.