Text to speech: add model download button + per-model download size

[niksedk]

Addresses #11507 — makes engine/model download size visible and adds a way to download a model without first clicking Generate speech from text.

Changes

• Model download button next to the Model combo, shown only for local per-model engines (Qwen3 TTS ×2, VibeVoice, IndexTTS, CosyVoice3, F5-TTS, VoxCPM2, Chatterbox). Cloud engines (ElevenLabs/Mistral) and single-model engines (Kokoro/OmniVoice) don't show it.
  • Model not installed → the normal download confirmation + download dialog (reuses IsEngineInstalled, the same path Generate/Test use — no duplicated download logic).
  • Model already installed → "Re-download {model}?" prompt, for repairing a corrupted/partial download.
• Per-model download size shown in the Model combo items, e.g. 1.7B (~1.6 GB).
• Model install-status (the combo dots) and size knowledge moved into the view model as a single source of truth, replacing the size string literals that were scattered across the per-engine download prompts.

Notes

• Sizes are shown only where the codebase already knew them (Qwen3, Chatterbox; VibeVoice/IndexTTS embed size in the model key). No sizes were invented for CosyVoice3/F5/VoxCPM2.
• Builds clean (0 warnings). Runtime UI verification on macOS was blocked by Terminal lacking Screen Recording/Accessibility permissions; the maintainer is testing the running app manually.

🤖 Generated with Claude Code