Add file_t cache to tokenizer

[dmed256]

We're storing file contents everytime there is an #include, as opposed to caching the file source