Fix detection of indentation in nested context.

In nested context the Tokenizer will helpfully insert multiple DETENT
token. So you need to while-pop.

Included a couple of extra test that verify this behavior.

IPython/core/inputsplitter.py

@@ -142,7 +142,7 b' def find_next_indent(code):'
         tokens.pop()
     if not tokens:
         return 0
-    if tokens[-1].type in {tokenize.DEDENT, tokenize.NEWLINE, tokenize.COMMENT}:
+    while (tokens[-1].type in {tokenize.DEDENT, tokenize.NEWLINE, tokenize.COMMENT}):
         tokens.pop()
 
     if tokens[-1].type == INCOMPLETE_STRING:

IPython/core/tests/test_inputsplitter.py

@@ -629,6 +629,9 b' indentation_samples = ['
      '     1,', 5),
     ('b = """123', 0),
     ('', 0),
+    ('def f():\n    pass', 0),
+    ('class Bar:\n    def f():\n        pass', 4),
+    ('class Bar:\n    def f():\n        raise', 4),
 ]
 
 def test_find_next_indent():