load save with pathlib.Path objects

For python >= 3.6, objects representing paths now have the option to have a dunder `__fspath__` (See PEP 519 for more info) to return the string (or bytes) representation of the path. Pathlib on 3.6 includes this method so we should check for it first. Then, for pathlib objects from the py 3.4-3.5x, `__fspath__` is not available, so we dip into the py3k module to get the current python version's (2 or 3) string method (`unicode` in this case). So overall we're allowing filename to be a path object, but ensuring that it will be converted to the string/bytes representation of the path before passing it on to `image_klass.from_filename`

My only reservations is that `nib.load` lowest entry point for this type of handling, because if some one really wants to call `from_filename` from an image_klass directly, they won't have this pathlib compatibility- however I think that `nib.load` is a very common entry point, which justifies the placement.

Author: CRiddler

nibabel/loadsave.py

@@ -17,7 +17,7 @@
 from .filebasedimages import ImageFileError
 from .imageclasses import all_image_classes
 from .arrayproxy import is_proxy
-from .py3k import FileNotFoundError
+from .py3k import FileNotFoundError, unicode
 from .deprecated import deprecate_with_version
 
 
@@ -36,6 +36,11 @@ def load(filename, **kwargs):
     img : ``SpatialImage``
        Image of guessed type
     '''
+    if hasattr(filename, '__fspath__'):
+        filename = filename.__fspath__()
+    else:
+        filename = unicode(filename)
+    
     if not op.exists(filename):
         raise FileNotFoundError("No such file: '%s'" % filename)
     sniff = None