Add support for row pitch in ImageView and ImageViewMut

[RunDevelopment]

Resolves #56

This PR adds support for row pitch in image views, which makes cropping and en/decoding image rectangles possible. This is quite useful, but was a pain to implement, because the assumption row_pitch == width * bytes_per_pixel as built into a lot of en/decoders. (It was such a pain that this PR is actually my third attempt at implementing this. But I'm quite happy that I did. The library is better for it.)

I tested the new row pitch changes by en/decoding random images/data with a both contiguous and non-contiguous views. If everything is implemented correctly, they'll produce the same result. This is done for all possible combinations of format and color (currently 73*12).

The performance impact of this feature is negligible. It's maybe 1% on average. Truth be told, there was a lot of noise in my benchmark timings, so 1% is my best guesstimate. However, I can say with certainty that there are no dramatic slowdowns.

Changes:

• Add a row_pitch: usize field to ImageView and ImageViewMut. This makes it possible to use ImageView* not just for whole (contiguous) images, but also for cropped sub-views of images.
• Added public cropped and is_contiguous functions to the view types.
• Changed the decode_rect and Decoder::read_surface_rect APIs to make use of row pitch support in ImageViewMut. I also updated DecodeError to remove 2 error variants that are no longer possible and to add 1 new one.
• Removed the Rect type and replaced it with an Offset type.
• Updated all encoder and decoder implementations to support row pitch. Despite adding functionality, some implementations actually got simpler, which is very nice.
• A few minor fly-by improvements.

[Copilot]

Pull Request Overview

This PR adds support for custom row pitches to ImageView and ImageViewMut, enabling encoded/decoded operations on cropped or non-contiguous sub-views. It refactors encoders and decoders to consume these views instead of raw buffers and updates tests to validate that contiguous vs. non‐contiguous workflows produce identical results.

• Introduce row_pitch in ImageView/ImageViewMut plus new_with(), is_contiguous(), cropped(), and row iterators.
• Refactor decoder/encoder APIs to use image views and new utilities (for_each_chunk, for_each_f32_rgba_rows).
• Update and add tests (test_row_pitch) to verify contiguous and non-contiguous decode/encode yield the same output.

Reviewed Changes

Copilot reviewed 23 out of 23 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/util.rs	Switched read_dds_rect_as_u8 to the new read_surface_rect(image.view_mut(), rect) API
tests/encode.rs	Replaced manual buffer/stride logic with read_surface + .cropped(); added test_row_pitch
tests/decode.rs	Updated decode_rect tests to use ImageViewMut; added a decoding test_row_pitch
src/lib.rs	Added row_pitch field, constructors (new_with), is_contiguous, cropped, and row iterators
src/error.rs	Removed legacy pitch/buffer errors; added UnexpectedRectSize
src/resize.rs	Adjusted resizing to handle non‐contiguous views via row iterators
src/split.rs	Refactored split_surface_into_lines to use iterator-based ranges and cropping
src/encode/write_util.rs	New helpers: for_each_f32_rgba_rows and for_each_chunk
src/encode/*	Updated all encoder modules to accept ImageView, use chunk/row utilities
src/decode/*	Updated all decoder modules to accept ImageViewMut, use offset abstraction and row iterators

Comments suppressed due to low confidence (2)

src/error.rs:127

• [nitpick] The message "Unexpected rectangle size" is vague—consider including the expected vs. actual sizes or clarifying that the image view size must match the rectangle size to aid debugging.

            DecodingError::UnexpectedRectSize => {

src/split.rs:6

• [nitpick] Returning Option<impl Iterator<...>> can obscure control flow and make callers less flexible. Consider returning a Vec<Range<u32>> directly (or Box<dyn Iterator<...>>) to simplify the code and improve readability.

fn split_surface_into_lines(

[RunDevelopment]

Not even bad. I didn't expect copilot's reviews to be very good, but it actually found a bug my tests didn't catch. Sure, the fix wasn't 100% correct, but just pointing out the problem is good enough.

As for the bug itself, I fixed it and changed the tests to cover the changed code paths.