Use tiktoken (or some other means) to limit documents instead of `byteSize`

Currently, Dryad uses a byte count (file size) to avoid passing too-large source files to ChatGPT for summarization. This is both inefficient and inaccurate. It would be best to know **exactly** how many tokens a given source file would be so we can more accurately limit prompt payloads based on the capability of whatever chat model we're using.