streaming voice stack

Omni-modal voice typing: multimodal understanding, MoE, and streaming text output

Omni-modal voice typing needs a split pipeline, streaming audio, and Mac-local context. A look inside Loqua's voice stack.