This is the second bug of this type that I've fixed since refactoring
this code, so clearly we need to do something differently. We rename
m_not_owning_ptr to m_owning_ptr and make sure it starts out false (so
that we never try to free a pointer if it wasn't assigned yet.) Then we
introduce two methods for setting the m_ptr field: own_ptr() which sets
m_owning_ptr, and share_ptr() which does not. We use these methods
consistently everywhere that we previously set m_ptr directly.

Closes: #240