Avoid cloning immutable decode inputs

[bghira]

Summary

• retain Python bytes storage with PyBackedBytes while decoding outside the GIL
• remove the unconditional encoded-buffer to_vec() copy
• preserve safe behavior for mutable bytearray inputs through PyO3-owned snapshot storage
• add decode coverage for both input types

Root cause

The GIL-release path cloned every encoded input before decode so the borrowed Python slice could cross the allow_threads boundary. PyO3 already provides a Send/Sync owned view that keeps immutable Python bytes alive without copying.

Performance

The avoided copy is proportional to encoded input size. A 32 MiB rejected-input probe fell from 0.500 ms to effectively decoder-call overhead (0.0005 ms), isolating the removed copy. For real images, decoder work still dominates; this is a smaller win than the resize/crop changes.

Validation

• bytes and bytearray decode regression tests
• cargo test --features opencv,simd (60 passed)
• Python test suite (70 passed)
• repository pre-commit hooks, including competitive performance benchmarks

[Copilot]

Pull request overview

This pull request optimizes the Python imdecode_py binding to avoid an unconditional clone of immutable encoded inputs when decoding outside the GIL, while adding regression coverage for both immutable and mutable input buffers.

Changes:

• Switch imdecode_py to accept pyo3::pybacked::PyBackedBytes and remove the unconditional to_vec() copy before allow_threads.
• Add a Python test to validate decoding works for both bytes and bytearray inputs.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
tests/test_python_bindings.py	Adds coverage ensuring imdecode_py works for both bytes and bytearray buffers.
src/python_bindings.rs	Removes the unconditional encoded-buffer clone by using PyBackedBytes across the allow_threads boundary.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.