Soften chest2err-score: add display temperature tau=3.0 (default)

Fix loader: correct CADAD signature, bundle sentence-segmentation collate + concept vocab, decode_greedy path verified end-to-end

Self-contained bundle: merged backbone + decoder + tokenizer; chest2vec_0.6b as base; K_total-only score; severity caveat

Honest fix: chest2err-score uses K_total (severity head not reliably trained in v0.1)

Add training-procedure section: GPT-4o-mini variant generation with anatomy + category labels

Reframe output as chest2err-score (exp(-K_w)) — GREEN-style single-number quality signal in (0,1]

Initial release: chest2err sentence-grounded error decoder (τ_b=+0.763, pairwise acc=0.958)

initial commit