Known limitations of the current implementation:

- documentation changes are incomplete
- there's a reference leak I haven't tracked down yet

The leak is most visible by running:

  ./python -m test -R3:3 test_importlib

However, you can also see it by running:

  ./python -X showrefcount

Importing the array or _testmultiphase modules, and
then deleting them from both sys.modules and the local
namespace shows significant increases in the total
number of active references each cycle. By contrast,
with _testcapi (which continues to use single-phase
initialisation) the global refcounts stabilise after
a couple of cycles.

cleaning other modules.  The sys and builtins modules are cleaned last.

repr(module) uses %R to format module name and filenames, instead of '%s' and
'%U', so surrogates from undecodable bytes in a filename (PEP 383) are escaped.

zero-initialize all fields, fixing compiler warnings seen when building
extension modules with gcc with "-Wmissing-field-initializers" (implied
by "-W")

... to get the filename as a unicode object, instead of a byte string. Function
needed to support unencodable filenames. Deprecate PyModule_GetFilename() in
favor on the new function.