Split stdlib into Python 2 and 3 versions (#5442)

All new files in stdlib/@python2 are straight copies of the
corresponding files in stdlib.

stdlib/@python2/unicodedata.pyi

@@ -0,0 +1,42 @@
+import sys
+from typing import Any, Text, TypeVar, Union
+
+ucd_3_2_0: UCD
+ucnhash_CAPI: Any
+unidata_version: str
+
+_T = TypeVar("_T")
+
+def bidirectional(__chr: Text) -> Text: ...
+def category(__chr: Text) -> Text: ...
+def combining(__chr: Text) -> int: ...
+def decimal(__chr: Text, __default: _T = ...) -> Union[int, _T]: ...
+def decomposition(__chr: Text) -> Text: ...
+def digit(__chr: Text, __default: _T = ...) -> Union[int, _T]: ...
+def east_asian_width(__chr: Text) -> Text: ...
+
+if sys.version_info >= (3, 8):
+    def is_normalized(__form: str, __unistr: str) -> bool: ...
+
+def lookup(__name: Union[Text, bytes]) -> Text: ...
+def mirrored(__chr: Text) -> int: ...
+def name(__chr: Text, __default: _T = ...) -> Union[Text, _T]: ...
+def normalize(__form: Text, __unistr: Text) -> Text: ...
+def numeric(__chr: Text, __default: _T = ...) -> Union[float, _T]: ...
+
+class UCD(object):
+    # The methods below are constructed from the same array in C
+    # (unicodedata_functions) and hence identical to the methods above.
+    unidata_version: str
+    def bidirectional(self, __chr: Text) -> str: ...
+    def category(self, __chr: Text) -> str: ...
+    def combining(self, __chr: Text) -> int: ...
+    def decimal(self, __chr: Text, __default: _T = ...) -> Union[int, _T]: ...
+    def decomposition(self, __chr: Text) -> str: ...
+    def digit(self, __chr: Text, __default: _T = ...) -> Union[int, _T]: ...
+    def east_asian_width(self, __chr: Text) -> str: ...
+    def lookup(self, __name: Union[Text, bytes]) -> Text: ...
+    def mirrored(self, __chr: Text) -> int: ...
+    def name(self, __chr: Text, __default: _T = ...) -> Union[Text, _T]: ...
+    def normalize(self, __form: Text, __unistr: Text) -> Text: ...
+    def numeric(self, __chr: Text, __default: _T = ...) -> Union[float, _T]: ...