feat: replace local Supertonic TTS with Voxtral

[AnthonyRonning]

Summary

• replace local Supertonic inference with encrypted OpenSecret/Voxtral synthesis while preserving existing desktop and iOS playback/cancellation behavior
• remove Supertonic models, dependencies, commands, download UI, and docs; delete exact legacy model directories at launch while retaining ONNX Runtime for PDF OCR
• normalize bullets, Markdown list markers, separator dashes, and other non-verbal text before synthesis without damaging negatives, hyphenated words, ranges, or expressions such as C++
• add Preferences controls for all 20 provider voice/accent presets and 0.5x–2.0x speech speed
• default new or reset preferences to Casual Female at 1.5x, while preserving existing valid ttsVoice and ttsPlaybackSpeed selections
• surface concise JSON provider errors before audio decoding while preserving cancellation behavior
• keep format, language, streaming, and advanced provider parameters out of Maple's user-facing settings

Dependency

• OpenSecret backend: Proxy paid Voxtral speech through Tinfoil opensecret#241

Maple uses published @opensecret/react 3.2.1 aiCustomFetch, so no SDK change or release is required. The backend must be deployed before remote synthesis works outside the managed workspace.

Validation

• rebased onto current master, retaining the new agent context-limit and Linux updater behavior alongside the TTS replacement
• product defaults selected after manual voice testing: Casual Female at 1.5x
• full frontend CI: formatting, TypeScript, ESLint with 0 errors, and 396 tests / 1,398 assertions
• focused TTS coverage: 37 tests / 88 assertions, including persisted former-default and Reset regressions
• full native suite: 209 passed, 1 model-backed OCR test ignored; all legacy-cleanup tests passed
• managed-stack transport/audio smoke performed earlier on the feature:
  • Free account rejected with HTTP 403
  • Pro generated valid neutral and French-accent PCM16 mono 24 kHz WAVs at multiple speeds
  • local Whisper transcribed every sample correctly or near-exactly
  • the exact bundled macOS app exposed the controls and completed Read aloud → Stop speaking → Read aloud
• no app was launched and no audio was played for the defaults-only follow-up; its runtime behavior is covered by focused preference/request tests
• git diff --check

Known gaps

• Physical iOS/TestFlight locked-device playback and Windows/Linux runtime playback have not been retested.
• Web and Android TTS remain intentionally unsupported.

[cloudflare-workers-and-pages]

Deploying maple with    Cloudflare Pages

Latest commit:	c24e5bd
Status:	✅  Deploy successful!
Preview URL:	https://b44eef57.maple-ca8.pages.dev
Branch Preview URL:	https://codex-maple-voxtral-tts-mapl.maple-ca8.pages.dev

View logs