Add test case for memory leak when using lambdas

[yeganer]

Using a function as an argument to a C++ object causes a reference cycle
that the garbage collector is somehow not able to break, causing the
corresponding C++ object to leak.

[rwgk]

I'd say there is a leak somewhere, but where?

I'd expect that local is automatically cleared when the C++ local variable goes out of scope, unless the local(s) passed via py::exec is/are kept alive elsewhere.

We know with a high degree of certainty that py::dict doesn't leak. Something that could easily be demonstrated by removing the py::exec call.

General: I do know that Py_Initialize and Py_Finalize should only be called once each in a process:

https://github.com/rwgk/stuff/blob/dc76a8713a2fdaf520893d96255da56fe8a401ab/noddy/embedded_noddy_main_run.c

Python does not fully clean up after itself. I believe it could maybe be better than it is, but a full cleanup seems infeasible. See "Bugs and caveats" here:

https://docs.python.org/3/c-api/init.html#c.Py_FinalizeEx

One way to get more certainty about the leak would be to reduce py::exec to the raw Python C API equivalent (seems pretty straightforward, if someone wants to try). The outcome could well be that pybind11 is fine and that the leak is in Python. Or not, idk.