Initializing character tokens, and the NSString objects they use, results in a relatively high overhead. The tokenizer
now accumulated all successive characters in on accumulator token until a non-character token is emitted.
This change reduces the execution time of the performance test on the local machine by 48% (reference to baseline)
Instead of loading and running the HTML5Lib test files manually, the Xcode test cases are generated
on start for all the HTML5Lib tokenizer tests. Also, the performance test is moved into its own class.
This setup shouldn't require any extra maintenance when HTML5Lib adds new tokenizer test cases.
iska