new explainability tools & document similarity

- faster shuffled()
- better csv reader (BOM)
- vectors supplied to Models can now only be dicts (not sets)
- ! better explainability functions cc(), explain() and counterfactual()
- new vector functions and distance metrics
- ! document similarity
- fast sep() for when tokenize() is too slow