Use a simple parser instead of relying on regular expressions

[holizz]

From what I can tell, mustache isn't a regular language, so I made pystache parse a context-free language.

This closes issue #36.

[holizz]

On trying to use this with actual data, the performance turns out to be abysmal. I will close this and open another when it's not useless.

[cjerdonek]

Thanks for taking a crack at this. Do you have any idea how the speed of what you did compares to the speed of the version of the parser at the HEAD of the spec-compliant branch? If possible, would you be able to compare using the same test data you used before? Superficially, it looks like they might take a similar approach.

I'm also interested in seeing how the HEAD of the spec-compliant branch compares with what we have now in the master/development branches -- to inform whether we should bring it in. Thanks if you can do this.