Don't terminate tokenization if stack size changes

Previously Python import blocks were not tokenizing correctly since
the loop was prematurely terminating when a match at the end of the line
was reached and no tokens were generated for it.

This approach was incorrect since the tokenizer may have just popped a rule
and another loop could possibly pop more rules.

Now this early termination is only performed if the stack size hasn't changed.

src/text-mate-grammar.coffee

@@ -146,7 +146,7 @@ class TextMateGrammar
 
         tokens.push(nextTokens...)
         position = tokensEndPosition
-        break if position is line.length and nextTokens.length is 0
+        break if position is line.length and nextTokens.length is 0 and ruleStack.length is previousRuleStackLength
 
       else # push filler token for unmatched text at end of line
         if position < line.length