test(02-07): rehearsal script; furigana at three layers, and a per-thread Sudachi tokenizer

feat(02-07): furigana controls, level picker, persistence; every transcript line rendered from tokens

feat(02-06): analyze/translate/language_info server functions, tokens on the directive, language warm-up

feat(02-05): analyze(text) - the canonical token record

feat(02-04): CTranslate2 + sentencepiece translator with warm-up

feat(02-03): JLPT word and kanji level derivation (N1+, name)

fix(02-03): rank headword precedence in the JMdict lookup

feat(02-02): committed reading-override table (人気 -> ひとけ before の)

feat(02-02): okurigana-aligned ruby spans

feat(02-02): mode-C morphemes to whole-word units (rules A1-A7)

feat(02-03): compact JMdict loader with ranked headword lookup

test(02-03): add failing JMdict lookup tests

feat(01-10): vendor the 3D runtime and take the CDN out of the render path

docs(01-08): mirror the credits, the flow-down notice and the honest scope in README

feat(01-08): wire the turn loop - text and voice turns, thinking, replay, slower, latency

feat(01-08): add the server turn - synthesis, timeline, per-request timings, directive

fix(01-05): say plainly what this build does and does not do

feat(01-05): write the Space manifest and the deployable Phase 1 shell

feat(01-06): add the viseme mapping table and banker's frame quantisation

test(01-03): enforce the transport seam statically and in a real browser

feat(01-03): add both transports, the standalone stage page and the Gradio component

feat(01-04): implement synthesize() and the shared voice dataclasses

chore(01-02): bootstrap Python 3.12 toolchain, Git LFS and package skeleton