Module pixel 

Pixel value codec — convert raw band bytes ↔ f64 for resampling math.

Every resampling kernel does its arithmetic in f64, regardless of the source dtype. This module is the bridge: read one pixel as f64, do the math, write back as the dtype’s raw bytes (clamping to the dtype’s valid range).

Why f64

• All numeric pixel types fit losslessly into f64 (even u32 and f32)
• Bilinear / cubic / lanczos kernels need fractional math anyway
• The per-pixel f64 ↔ raw cost is dominated by the kernel evaluation, not the conversion

Functions

read

Read one pixel from a band at (col, row) as f64. Returns None if the band’s dtype isn’t supported by the resamplers (Rgb8/Rgba8 need per-channel handling that the caller does manually) or if the coordinate is out of range.

write

Write value into a band at (col, row). Clamps to the dtype’s valid range; rounds-to-nearest for integer types. Returns false if the coordinate is out of range OR the dtype isn’t supported.