model.ts: rename ONNX input ref_s → style for kokoro-js / kokoro-onnx interop

ui: group voice picker by gender via <optgroup> (24 voices was unwieldy as flat list)

synthesize: strip BOS+EOS audio from output (keep BOS in model input)

synthesize: prepend BOS/EOS to inputIds (was missing — caused leading-consonant eat)

audio: prepend 80ms leading silence to fix SpringLab utterance-initial consonant eat

punctuation pauses: splice silence after , ; : — . ! ? …

metrics: TTFB (time to first audio) replaces total-synth in headline metric

ui: performance metrics card + voice ref-clip download + WAV download

speed slider now scales pred_dur (was playback-rate only)

feat: per-word display shows user's original Roman input for substituted loanwords

cleanup: remove verification marker

fix: actually deploy textarea-preserving build (build entry was contaminated by previous dist)

src: align main.ts source with deploy (in-memory transliterate)

loanword preprocessor + re-exported v0.2 ONNX (torch 2.6)

v0.2 polish: per-voice default speeds + trailing audio pad

voicepacks: gender-paired chips + dynamic textarea default

release: webgpu demo + source