Vec<u8>: support u8-compatible buffer exporters

[espressolee]

Summary

This PR extends Vec<u8> extraction to support u8-compatible buffer-protocol exporters (e.g. memoryview, array('B'), and custom types implementing __getbuffer__) by performing a contiguous copy via PyBuffer<u8>.

When the object exports a buffer that is not compatible with u8 (e.g. array('I')), extraction falls back to the existing sequence semantics, preserving current behavior.

Why

Vec<u8> is a common "byte payload" type. Today:

• bytes / bytearray already have a specialized path via u8::sequence_extractor.
• Other buffer exporters can still hit element-wise sequence extraction, and non-sequence buffer exporters cannot be extracted into Vec<u8> at all.

This change treats the buffer protocol as a first-class source for Vec<u8> when it is semantically valid (u8-compatible).

Implementation notes

• Extends the u8::sequence_extractor specialization to recognize buffer exporters.
• Makes the internal FromPyObjectSequence::to_vec fallible (PyResult<Vec<_>>) so buffer copies can propagate errors cleanly.
• Uses the existing PyBuffer<u8> implementation to perform the copy.

Compatibility

• No change to str -> Vec<_> rejection.
• bytes / bytearray behavior unchanged.
• For incompatible buffers (e.g. array('I')), behavior is preserved via fallback to sequence semantics.

Tests

Added to tests/test_buffer_protocol.rs:

• test_extract_vec_u8_from_buffer_exporter (custom buffer exporter, not a sequence)
• test_extract_vec_u8_falls_back_when_buffer_incompatible (array('I') fallback)

[codspeed-hq]

CodSpeed Performance Report

Merging this PR will degrade performance by 11.58%

_{Comparing espressolee:vec-u8-buffer-fast-path (b734b6a) with main (dfd6eaf)}

Summary

❌ 1 regressed benchmark
✅ 98 untouched benchmarks
⏩ 1 skipped benchmark¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	vec_bytes_from_py_bytes_medium	2 µs	2.3 µs	-11.58%

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩

[davidhewitt]

I am unsure I am comfortable with having this be built-in to the Vec<u8> extraction - see https://alexgaynor.net/2022/oct/23/buffers-on-the-edge/ - there is no guarantee that the buffer object is properly synchronized. By leaving this to user code we let them make the choice.

Hopefully the Python C API will one day support better guarantees for buffer objects.