deploy: main@5d35a5b (batched narration read-path: clip offsets + clipRef headers so the reader can play ONE sentence out of a shared blob; + research agent)

deploy: real page counts in /audio/voices + ratings

deploy: /audio/voices per-book voice status + Supertonic provider

deploy: audiobook subsystem + encode int16 fix + R2 page fast-path + costs breakdown + query-run flow

feat(audiobook): additive TTS generation + serving (first slice)