Remove the remove_newline hack and build that stuff into error recovery instead.

Tests are passing except for diff parser tests.

test/test_parser.py

@@ -81,7 +81,7 @@ def test_incomplete_list_comprehension():
     # parser only valid statements generate one.
     children = parse('(1 for def').children
     assert [c.type for c in children] == \
-        ['error_node', 'error_node', 'newline', 'endmarker']
+        ['error_node', 'error_node', 'endmarker']
 
 
 def test_newline_positions():
@@ -153,7 +153,7 @@ def test_python2_octal():
 def test_python3_octal():
     module = parse('0o660')
     if py_version >= 30:
-        assert module.children[0].children[0].type == 'number'
+        assert module.children[0].type == 'number'
     else:
         assert module.children[0].type == 'error_node'
 

test/test_parser_tree.py

@@ -63,7 +63,7 @@ class TestsFunctionAndLambdaParsing(object):
 
 def test_end_pos_line():
     # jedi issue #150
-    s = "x()\nx( )\nx(  )\nx (  )"
+    s = "x()\nx( )\nx(  )\nx (  )\n"
     module = parse(s)
     for i, simple_stmt in enumerate(module.children[:-1]):
         expr_stmt = simple_stmt.children[0]

test/test_prefix.py

@@ -12,7 +12,6 @@ import parso
     (' \f ', [' ', '\f', ' ']),
     (' \f ', [' ', '\f', ' ']),
     (' \r\n', [' ', '\r\n']),
-    ('\\', ['\\']),
     ('\\\n', ['\\\n']),
     ('\\\r\n', ['\\\r\n']),
     ('\t\t\n\t', ['\t\t', '\n', '\t']),
@@ -43,7 +42,6 @@ def test_simple_prefix_splitting(string, tokens):
     ('\r\n', ['newline']),
     ('\f', ['formfeed']),
     ('\\\n', ['backslash']),
-    ('\r', ['newline']),
 ])
 def test_prefix_splitting_types(string, types):
     tree = parso.parse(string)