a2C vocabulary: refit on the cleaned, budget-proportional corpus

docs: drop an unsupported claim

docs: correct the fertility delta and spell out the eviction example

a2B: reproducible seed, public benchmarks, benchmark.py

a2u: 28,000-stem seed, measured on text withheld from the fit

docs: the diminutive joke, as the counterexample for over-splitting

picur-tokenizer a2x: 49,152-piece Hungarian Unigram vocabulary

initial commit