Fix a more bytes/str confusion.

Use str.encode('raw-unicode-escape') consistently instead of bytes(string).

Remove the convert_eols argument from base64mime.decode().  This matches
previous API changes done to the quoprimime module.
diff --git a/Lib/email/test/test_email.py b/Lib/email/test/test_email.py
index fc8224a..981441c 100644
--- a/Lib/email/test/test_email.py
+++ b/Lib/email/test/test_email.py
@@ -2531,10 +2531,8 @@
 
     def test_decode(self):
         eq = self.assertEqual
-        eq(base64mime.decode(''), '')
+        eq(base64mime.decode(''), b'')
         eq(base64mime.decode('aGVsbG8='), b'hello')
-        eq(base64mime.decode('aGVsbG8=', 'X'), b'hello')
-        eq(base64mime.decode('aGVsbG8NCndvcmxk\n', 'X'), b'helloXworld')
 
     def test_encode(self):
         eq = self.assertEqual
@@ -2844,7 +2842,7 @@
     def test_empty_header_encode(self):
         h = Header()
         self.assertEqual(h.encode(), '')
-        
+
     def test_header_ctor_default_args(self):
         eq = self.ndiffAssertEqual
         h = Header()