Markdown parser performance improvement.

Do a quick lookahead to make sure what follows looks like a setext
header before parsing any Inlines.  This gives a 15% performance
boost in one benchmark.  Many thanks to knieriem for finding
the problem (in peg-markdown):

https://github.com/jgm/peg-markdown/issues/issue/3

src/Text/Pandoc/Readers/Markdown.hs

@@ -320,6 +320,9 @@ atxClosing = try $ skipMany (char '#') >> blanklines
 
 setextHeader :: GenParser Char ParserState Block
 setextHeader = try $ do
+  -- This lookahead prevents us from wasting time parsing Inlines
+  -- unless necessary -- it gives a significant performance boost.
+  lookAhead $ anyLine >> many1 (oneOf setextHChars) >> blankline
   text <- many1Till inline newline
   underlineChar <- oneOf setextHChars
   many (char underlineChar)