feat: add Windows CUDA GPU model support with NSIS installer

[gcmullen]

Add GPU-accelerated Parakeet model support on Windows via NVIDIA CUDA, with runtime detection, bundled ONNX Runtime DLLs, an NSIS installer with prerequisite checks, and a rewritten model management system.

Model management:

• Overhaul model.rs with CUDA detection (CUDA_PATH/CUDNN_PATH env vars)
• Add HuggingFace multi-file download support for GPU model variants
• Add Whisper Small/Medium/Turbo/Large, Moonshine Base, and Parakeet V2/V3 in both CPU (INT8) and GPU (FP32) variants
• GPU models only visible when CUDA + cuDNN are detected at runtime

Build and installer:

• Add NSIS installer hooks checking for CUDA/cuDNN prerequisites
• Add build.rs logic to copy ONNX Runtime DLLs for Windows builds
• Add CI step to download ONNX Runtime GPU DLLs during workflow
• Bundle onnxruntime*.dll in resources/gpu-deps (DLLs gitignored)
• Change Windows build target to NSIS-only

Runtime:

• Setup GPU DLL PATH in lib.rs for ONNX Runtime + CUDA discovery
• Update transcription manager for CUDA-aware engine loading

Dependency changes:

• Update transcribe-rs to use git fork with CUDA support
• Dependency bumps (tauri, tauri-apps packages)
• Frontend updates for new model variants

Also includes: documentation and changelog updates.

Cleanup:

• Remove upstream-specific GitHub templates (FUNDING, issue/PR templates)