Rework Yaml constructor/representer to match PyYaml's setup

They apparently should be called on class objects rather than on instances.
I ran into thread-safety problems (eating data) before this change.

/cc http://pyyaml.org/ticket/36 which mentions those methods not being
thread-safe, but that PyYaml itself should be.

octodns/yaml.py

@@ -30,11 +30,7 @@ def _zero_padded_numbers(s):
 # here
 class SortEnforcingLoader(SafeLoader):
 
-    def __init__(self, *args, **kwargs):
-        super(SortEnforcingLoader, self).__init__(*args, **kwargs)
-        self.add_constructor(self.DEFAULT_MAPPING_TAG, self._construct)
-
-    def _construct(self, _, node):
+    def _construct(self, node):
         self.flatten_mapping(node)
         ret = self.construct_pairs(node)
         keys = [d[0] for d in ret]
@@ -44,6 +40,10 @@ class SortEnforcingLoader(SafeLoader):
         return dict(ret)
 
 
+SortEnforcingLoader.add_constructor(SortEnforcingLoader.DEFAULT_MAPPING_TAG,
+                                    SortEnforcingLoader._construct)
+
+
 def safe_load(stream, enforce_order=True):
     return load(stream, SortEnforcingLoader if enforce_order else SafeLoader)
 
@@ -57,16 +57,15 @@ class SortingDumper(SafeDumper):
     more info
     '''
 
-    def __init__(self, *args, **kwargs):
-        super(SortingDumper, self).__init__(*args, **kwargs)
-        self.add_representer(dict, self._representer)
-
-    def _representer(self, _, data):
+    def _representer(self, data):
         data = data.items()
         data.sort(key=lambda d: _zero_padded_numbers(d[0]))
         return self.represent_mapping(self.DEFAULT_MAPPING_TAG, data)
 
 
+SortingDumper.add_representer(dict, SortingDumper._representer)
+
+
 def safe_dump(data, fh, **options):
     kwargs = {
         'canonical': False,