rewrite backend

[stephantul]

This PR completely rewrites the distill backend.

• There's no longer a difference between subword and vocabulary inference
• BPE and unigram tokenizer token removal/addition is supported
• All tokenizers now take into account normalization when adding tokens
• Tokens that are multi-word units are no longer allowed
• Make tokenizers a bit smaller by removing unneeded special tokens

[codecov]

Codecov Report

Attention: Patch coverage is 76.30058% with 41 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
model2vec/distill/tokenizer.py	47.05%	36 Missing ⚠️
model2vec/distill/distillation.py	91.66%	3 Missing ⚠️
model2vec/distill/inference.py	95.55%	2 Missing ⚠️

Files with missing lines	Coverage Δ
model2vec/distill/utils.py	100.00% <100.00%> (ø)
tests/conftest.py	100.00% <100.00%> (ø)
tests/test_distillation.py	89.88% <100.00%> (-0.12%)	⬇️
model2vec/distill/inference.py	95.58% <95.55%> (+1.74%)	⬆️
model2vec/distill/distillation.py	94.73% <91.66%> (-1.10%)	⬇️
model2vec/distill/tokenizer.py	51.35% <47.05%> (-24.52%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Pringled]

BIG CLEAN 🧹 few small concerns