refactor(core): overhaul audio engine and device architecture

[LSXPrime]

This commit introduces a major architectural refactoring to move away from a singleton-based (AudioEngine.Instance) model to a device-centric, instance-based architecture. This fundamental change enables true multi-device input and output, allowing applications to manage and switch between multiple playback and capture devices simultaneously and independently.

This refactor also includes several new features, performance improvements, and critical bug fixes that are built upon the new architecture.

Key Changes & Features

• Multi-Device Architecture: The global AudioEngine.Instance and Mixer.Master have been removed. The new design introduces AudioPlaybackDevice, AudioCaptureDevice, and FullDuplexDevice abstractions, each managing its own audio graph and MasterMixer.
• Explicit Context: Core components now receive AudioEngine and AudioFormat context via their constructors, removing implicit dependencies on global state.
• Device Switching API: A new, robust Engine.SwitchDevice() API allows for seamless hot-swapping of active audio devices while preserving their state (e.g., attached components or event subscribers).
• New VoiceIsolationEffect: Adds an experimental modifier that acts as a band-pass filter to help isolate human speech frequencies.
• New MultiEngines Sample: A new sample project is included to demonstrate and test the new multi-device capabilities.

Fixes & Performance Improvements

• fix(network): Refactored NetworkDataProvider to use a chunked streaming and decoding model, resolving excessive memory usage. Memory consumption for a 15-minute audio file is reduced from ~1GB to ~20MB.
• fix(recording): Corrects a critical audio format conversion bug that produced garbled audio when recording to formats other than F32.
• refactor(native): Unified the FFT implementations in MathHelper for Scalar, SSE, and AVX paths to ensure consistent behavior. The SIMD paths are now toggleable via static properties for easier debugging.
• refactor(interop): The native device configuration layer now uses a comprehensive DTO, allowing for more granular control over backend-specific settings (WASAPI, ALSA, etc.).

BREAKING CHANGE: This is a fundamental architectural change that affects the entire public API.

• AudioEngine.Instance is removed. An AudioEngine must now be instantiated (e.g., new MiniAudioEngine()).
• Audio is no longer played by adding components to Mixer.Master. Users must first initialize a device (e.g., engine.InitializePlaybackDevice(...)) and then add components to the device's MasterMixer.
• Constructors for nearly all SoundComponent, ISoundDataProvider, SoundModifier, and AudioAnalyzer derivatives have changed. They now require an AudioEngine and/or AudioFormat instance to be passed in.
• The global AudioEngine.OnAudioProcessed event is removed. Events for processed audio are now available on a per-device basis (e.g., AudioCaptureDevice.OnAudioProcessed).
• The previous SwitchDevice API is replaced with new, more explicit methods on the AudioEngine instance that return a new device object.