feat: Voxlen API settings, context selector, SDK rewire #4773 — ccantynz/voxlen

Originally written by @ccantynz-alt on GitHub.
Imported from https://github.com/ccantynz-alt/voxlen/pull/28

Summary

Voxlen API integration — SDK rewired to route through api.voxlen.com/v1 when voxlenApiKey is set; new VoxlenApiClient with transcribe, async transcribe, SSE stream, vocabulary CRUD, health check
Settings — Voxlen API tab — new tab with API key (masked), tenant ID, Test Connection button, context dropdown (13 domain contexts), Privileged Mode, Legal Mode, Jurisdiction, Billable Rate
Context selector in DictationPanel — quick-switch dictation context without opening settings; compact pill dropdown mid-session
11 context formatters (src/lib/contextFormat.ts) — legal contract, case note, court filing, deposition, correspondence; accounting tax, audit, memo, correspondence
Settings store extended — voxlenApiKey, voxlenContext, voxlenTenantId persisted
Grammar upgraded — Haiku → claude-sonnet-4-6 in SDK
Deepgram nova-3 in SDK, utterance_end_ms=1000, no_delay=true
Fix speech-recognition.d.ts duplicate index signatures (4 TS errors)

Settings > Voxlen API — enter key, click Test Connection → "Connected" or error
Change Dictation Context in dropdown — voxlenContext store value updates immediately
Dictate with context set to legal_contract — defined terms capitalised in output
Dictate with accounting_tax — GST/VAT/IRD references formatted correctly
Privileged Mode toggle → shield shows in TitleBar, dictation falls back to local STT
Legal Mode on → Latin phrases (inter alia, res judicata) formatted correctly
tsc --noEmit passes with zero errors

Generated by Claude Code