I don't think we should go down this path.  The efficiency of the C implementation depends on it being tightly coupled to lists.  This tool is used in the schedulers of various async tools (such as Tornando), used for merge(), nsmallest(), and nlargest() all of which depend on this foundational tool being very fast.

Also, I question whether it makes sense at all to be heapifying numpy arrays using standard library tooling.  It numpy arrays actually needed this and needed for it to be efficient, it would need to be implemented natively in numpy.