perf: optimize audio processing hot paths (mixer, transformers, buffer pool, worker IPC)

[quiverdev]

Changes

This PR implements a series of high-impact performance optimizations targeting the hot paths of NodeLink's audio processing engine:

1. Audio Mixing – Refactored the inner loop of AudioMixer.mixBuffers to pre-filter active layers and use local array references, significantly reducing CPU cycles spent on property lookups and conditional logic.
2. Transformers – Added specialized logic to VolumeTransformer and FadeTransformer to skip expensive calculations when gain remains constant between frames.
3. Math Optimizations – Replaced division operations with multiplications by pre-calculated inverses in LoudnessNormalizer and implemented a Taylor series approximation for the soft-knee limiter in VolumeTransformer.
4. Memory Management – Increased default BufferPool capacity to reduce allocation frequency under load. Added RingBuffer.readTo to allow direct copying into existing buffers, avoiding unnecessary Buffer object allocations.
5. Worker IPC – Rounded floating-point numbers in the worker statistics payload to decrease the size of serialized data transmitted between processes.

Why

These optimisations collectively improve both CPU throughput and RAM stability by reducing garbage collection pressure and instruction counts in real-time audio streams. The changes make NodeLink more efficient under high concurrency and heavy audio load, which is critical for production deployments.

Checkmarks

• The modified endpoints have been tested.
• Used the same indentation as the rest of the project.
• Still compatible with LavaLink clients.

Additional information

No additional information at this time.

[github-actions]

All Contributors have signed the CLA. The PR is now allowed to be merged.
_{Posted by the CLA Assistant Lite bot.}

[quiverdev]

I have read the CLA Document and I hereby sign the CLA