Add support for cross-encoder models (+fix token type ids) (#501)
* Formatting
* Update ESM pair template
* Fix token type ids
* Update JSDoc
* Cleanup
* Remove unused `prepare_model_inputs` function
* Move pad and truncate logic to helper functions
* Add static padding/truncation unit tests
* Fix padding/truncation
* Remove unused `add_token_types` function
* Reduce duplication
* `let` -> `const` where possible
* Add cross-encoder models