feat(speech): speech to prompt (dictation)

[n0tlu5]

Refs #844

First PR of the speech stack for #844; it targets main directly.

What

Dictation for the composer. A mic button records speech and inserts the text into the prompt, using either the browser's speech recognition or an OpenAI-compatible transcription endpoint. It also adds a Speech settings page (engine, language, endpoint and API key; the key never leaves the backend) and microphone entitlements for the desktop app. On plain-http web deployments, browser dictation is reported as unavailable instead of failing on the first click; the cloud engine keeps working.

Commits to review

• feat(speech): add speech preferences and capability detection
• feat(speech): add OpenAI-compatible cloud transcription backend
• feat(desktop): allow microphone capture in app windows
• feat(chat): add speech input hook with browser and cloud engines
• feat(chat): add voice dictation button to the composer
• feat(settings): add Speech settings page
• fix(speech): report browser dictation unavailable on insecure origins

Checks

pnpm lint ., pnpm test, pnpm build; desktop cargo check, cargo clippy --all-targets --features test-utils -- -D warnings and cargo test --features test-utils; server cargo check, cargo test --lib and cargo clippy -D warnings; codeg-mcp cargo check and clippy. All pass at the tip of the stack, and the behaviour was exercised in web mode in Chromium. No new npm or cargo dependencies.