zignal 0.7.1

Highlights

Faster pipelines: Convolution now skips redundant work for uniform RGB channels and leverages SIMD kernels, while the sixel encoder received a full rewrite with smarter palette generation, chunking, and profiling hooks, dramatically cutting time-to-frame for terminal renders.
Safer numerics: Matrix GEMM now correctly dispatches Aᵀ * Bᵀ paths to SIMD helpers, and JPEG/PNG decoders gained hardened restart-marker handling plus accurate 16‑bit channel extraction to stop stray color swaps.
Better accuracy: Feature Distribution Matching honours grayscale targets again, rectangles perform more stable overlap checks, and canvas drawing floors fractional coordinates to eliminate off-by-one artefacts.
Tidier internals: Image quality helpers (PSNR/SSIM) moved into image/metrics.zig, paving the way for new example (like the fresh metrics demo) while web samples were refreshed to showcase the improved terminal encoders.

Added zignal.RunningStats, mirroring Zig’s streaming statistics (mean, variance, skewness, etc.) with add, extend, combine, and scale helpers plus read-only properties.
Exposed zignal.perlin(...), bringing the 3D Perlin noise generator to Python with guarded amplitude, frequency, octave, persistence and lacunarity controls.

Faster pipelines: Convolution now skips redundant work for uniform RGB channels and leverages SIMD kernels, while the sixel encoder received a full rewrite with smarter palette generation, chunking, and profiling hooks, dramatically cutting time-to-frame for terminal renders.
Safer numerics: Matrix GEMM now correctly dispatches Aᵀ * Bᵀ paths to SIMD helpers, and JPEG/PNG decoders gained hardened restart-marker handling plus accurate 16‑bit channel extraction to stop stray color swaps.
Better accuracy: Feature Distribution Matching honours grayscale targets again, rectangles perform more stable overlap checks, and canvas drawing floors fractional coordinates to eliminate off-by-one artefacts.
Tidier internals: Image quality helpers (PSNR/SSIM) moved into image/metrics.zig, paving the way for new example (like the fresh metrics demo) while web samples were refreshed to showcase the improved terminal encoders.

Added zignal.RunningStats, mirroring Zig’s streaming statistics (mean, variance, skewness, etc.) with add, extend, combine, and scale helpers plus read-only properties.
Exposed zignal.perlin(...), bringing the 3D Perlin noise generator to Python with guarded amplitude, frequency, octave, persistence and lacunarity controls.