Wave v0.4.0

The pipeline is 400 to 800ms faster. A short utterance can now finish in
under a second, end to end. Typical was 1.8 to 5.3 seconds and is now 1.0 to
4.5, with a best case around 0.8 to 1.2.

Streaming API responses. The Claude and GPT calls use SSE streaming rather
than waiting for the whole HTTP response, so text arrives 200 to 500ms sooner.

Prefetching during recording. The dictionary, the snippets and the rewrite
context load while you are still talking rather than after you stop.

Audio work moved to Accelerate. Resampling, the RMS calculation and the
float to int16 conversion use vDSP instead of manual loops.

A faster paste. The internal delays went from 200ms to 70ms, and prompt
templates load from disk once and then stay in memory. API calls carry a ten
second timeout, so a hang ends rather than lasting forever.

Open Wave.dmg and drag Wave into Applications. This build is signed ad hoc
rather than with a Developer ID, so Gatekeeper refuses the first launch: open
System Settings, Privacy & Security, and the note about Wave near the bottom has
an Open Anyway button. Requires macOS 14 or later, Apple Silicon, and an
Anthropic or OpenAI API key.

sha256 89c16a3ce74acf4e525771f593837ba61037432656924b7d0edf018219441866