Removed cache* functions altogether. mustache.parse is all that's
really needed.

This commit allows pre-compiled templates to be executed without
keeping the original template around. It is still required when using
higher-order functions.

Fixes #262

Fixes #247

Text tokens now accurately reflect their content's exact position
in the original template.

Fixes #207 #210

Also, allow partials to be passed to the function that is returned
for pre-compiled templates. This makes the API more consistent all
around, but comes with the caveat that renderers are much more
aggressive at caching. Now, any time a template or partial is
compiled with the same name as a previously compiled template the
cached version is used.

Fixes #240

Fixes #187