cua_s1: preprocess RGB images for native vision on CPU

[Levius-Fubuki]

Purpose

Add image_preprocess::preprocess_rgb8(width, height, rgb) for decoded interleaved RGB8 input. Return float32 pixel patches, grid, resized dimensions and image-token count using the fixed Qwen3.5-4B processor: smart resize, uint8 antialiased bicubic filtering, fused normalization, merge ordering and temporal repetition. Validate dimensions, aspect ratio and exact input length before image allocations.

Updated onto main 7f39ac40902c374803992407bb26eeba29c8a588. The six-file diff contains the standalone CPU implementation, its export and the required upstream notices/licenses. No runtime dependency is added. Decoder/HTTP wiring and CUDA vision execution are supplied separately by #64.

Test Plan

Current head: 274b082b29d34a73cb0b6a48cb573a540fe732bc. Baseline: 7f39ac40902c374803992407bb26eeba29c8a588.

Test Result (fresh, 2026-10-05)

The current production tree passed cargo fmt --all --check, cargo clippy --workspace --locked --all-targets -- -D warnings, cargo test --workspace --locked, and cargo build --workspace --release --locked on macOS arm64 / Rust 1.98.0. Workspace tests: 82 passed, 0 failed, 10 existing tests ignored (6 CUDA, 2 Laya checkpoint/oracle, 1 Laya packing, 1 Open-Jev tokenization). Benchmark CI checks also passed: 4 smoke requests validated and 8 Python tests passed.

Feature fixtures and examples remain outside this core-only diff. A fresh external Cargo runner imports the current production crate by path and uses its locked dependency versions; only the external root package is added to the runner lockfile. The feature harness is materialized from the linked pre-cleanup commit. No native CUDA, full multimodal inference, HTTP image path, model accuracy or performance claim is made by this CPU prerequisite PR.

The external release suite passed all 7 tests, including input rejection, inclusive limits, identity patch order, temporal repetition, normalization, Python ties-to-even geometry and all 14 complete-output golden hashes. The unchanged fixed 24-case held-out corpus was replayed against the actual CPU Hugging Face processor: all 197,468,160 float32 output bytes matched exactly, plus grid, shape, resized dimensions and token counts. Zero tolerance; no resampling or threshold changes.

Reference: processor config revision 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a, config SHA-256 27225450ac9c6529872ee1924fcb0962ff5634834f817040f444118116f4e516. Reference versions were checked before replay: Python 3.12, Torch 2.14.0, torchvision 0.29.0, Transformers 5.17.0, NumPy 2.5.3, Pillow 11.3.0. No model weights or GPU resources were used.

Harness sources: pre-cleanup commit. Fresh local logs, runner lockfile, frozen protocol, full output hashes and held-out results are archived under artifacts/all-pr-followups-20261005/cua/pr63/.

Self-review

Full current diff reviewed against the pinned main revision; model-specific preprocessing remains separate from forward execution. Required PyTorch, Pillow and Apache notices remain with the adaptation. This CPU preprocessing module can merge independently as a prerequisite for #64. Maintainer review remains separate.

Demo / evidence

Actual CPU validation, not native encoder/HTTP inference. Raw logs and fixture/payload hashes remain local under the artifact path above, outside the core-code diff. No video is needed for these CPU prerequisites. Current-head GitHub Rust, benchmark and docs CI passed; deployment skipped. CI run.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.