Feature Request: Add Whisper Large V3 Turbo model

## Problem

Whisper Large V3 provides better accuracy than Parakeet for non-native English speakers (especially punctuation placement), but it's very slow. Whisper Large V3 Turbo offers the same accuracy with ~8x faster inference (4 decoder layers vs 32).

Issue #111 requested this but was closed after Parakeet was added. However, Parakeet doesn't solve the accuracy problem for accented speech with natural thinking pauses — it places punctuation incorrectly when the speaker pauses mid-thought, and sometimes transcribes wrong words.

## Requested solution

Add `openai/whisper-large-v3-turbo` as a model option in Settings → Transcription Model, alongside the existing Whisper Large V3 and Parakeet options.

The CoreML-converted model is available via WhisperKit: https://huggingface.co/argmaxinc/whisperkit-coreml/tree/main (look for `openai_whisper-large-v3_turbo` variants).

## Why this matters

- **Non-native speakers** get better accuracy from Whisper than Parakeet
- **Large V3 is too slow** for real-time dictation use
- **Large V3 Turbo** is the sweet spot: Whisper-level accuracy at near-Parakeet speed
- The model is already CoreML-compatible via WhisperKit

## Environment

- macOS 15 Sequoia, Apple Silicon (M2 / M4)
- Hex version: latest

Start in Settings → Transcription Model and inspect how the existing Whisper Large V3 and Parakeet options are registered. Check the linked WhisperKit CoreML repository for the openai_whisper-large-v3_turbo variants and verify the selected model can be used for transcription. Done means the new option appears alongside the existing models and works for dictation.

Written by the indexing model from the issue text.