docs: correct v3 speed claim — gains scale with video length, not a fixed multiplier

docs: add v3 to README — pipelined stages, ~6m45s for 2:39 video on CPU

fix: wrap run_in_executor in async def for create_task compatibility

v3: pipelined transcribe → translate → synthesize concurrently

docs: clarify v2 focus on speed (~2x faster), not quality

v2: synthesis optimisations — embedding cache + segment merging + smaller Whisper

docs: add README with setup, architecture and version history

v1: YouTube voice-cloning translator