This PR addresses two memory issues related to video capture, the first of which was identified by the Unity Profiler as a key contributor to broader performance problems:

1. Extraneous allocations in WebCameraSource: the read buffer method was performing two unnecessary heap allocations per call, leading to large GC events and severe performance impacts in other parts of the SDK (namely in audio capture):

// This results in 2 * width * heigh * 32 bytes being allocated on each invocation
Color32[] pixels = new Color32[GetWidth() * GetHeight()]; // first allocation
Texture.GetPixels32(pixels);
var bytes = MemoryMarshal.Cast<Color32, byte>(pixels);
_data.CopyFrom(bytes.ToArray()); // second allocation*

*Span<T>.ToArray also performs heap allocation as per the docs.

The solution is to only allocate the temporary buffer once until the configuration is invalidated, and use MemoryMarshal and Span methods to avoid the additional ToArray call:

CamTexture.GetPixels32(pixels);
MemoryMarshal.Cast<Color32, byte>(pixels)
    .CopyTo(_captureBuffer.AsSpan());

2. Native array used by RtcVideoSource and its subclasses was not disposed of, causing a memory leak; when a native array is created using Allocator.Persistent, a manual call to dispose is required. RtcVideoSource now implements IDispose to ensure the native array is disposed of even if Dispose() is not manually called.

CLT-1613