Fix confusing ReadError traceback

Avoids confusing tracebacks when ReadErrors exceptions are raised
from the yaml reader.

Now we do print the filename that failed to load and stop processing
in order to avoid additional exceptions.

Also adds tests to prevent future regressions.

Fixes: #261

tests/test_syntax_errors.py

@@ -92,3 +92,28 @@ class YamlLintTestCase(RuleTestCase):
                    '    a:\n'
                    '      set\n'
                    '...\n', None)
+
+    def test_invalid_char_last(self):
+        self.check('---\nkey: value\t\n',
+                   None, problem=(2, 11))
+
+    def test_invalid_char_first(self):
+        self.check('---\n\tkey: value\n',
+                   None, problem=(2, 1))
+
+    def test_invalid_char_on_second_line(self):
+        self.check('---\nfoo: bar\n\toutput: foo\n',
+                   None, problem=(3, 1))
+
+    def test_invalid_chars(self):
+        # We expect to identify only the first syntax error as this kind of
+        # error is fatal (stops parsing)
+        self.check('---\nkey: \tfoo\tbar\n',
+                   None,
+                   problem=(2, 6))
+
+    def test_invalid_char_with_other_rule(self):
+        self.check('key: value\t\n',
+                   None,
+                   problem1=(1, 1, 'document-start'),
+                   problem2=(1, 11))