release: v1.0.32

[justrach]

Summary

Prepare TurboAPI v1.0.32 for release.

This release includes the merged Linux ARM64 CPython 3.14t wheel work and its follow-up fixes:

• build and audit a manylinux_2_28_aarch64 wheel;
• provide the wheel-only getrandom compatibility implementation required by Zig 0.16 without exposing arc4random_buf;
• exercise the installed native backend with a real HTTP request in CI;
• document and test the Apple container workflow;
• remove per-request DHI model_dump closure patching while retaining list-field compatibility.

Version changes

• .bumpversion.cfg: 1.0.31 → 1.0.32
• pyproject.toml: 1.0.31 → 1.0.32
• python/setup.py: 1.0.31 → 1.0.32
• python/turboapi/__init__.py: 1.0.31 → 1.0.32

Local verification

• python -m pytest tests/test_release_version_sync.py tests/test_model_dump_compat.py -q
• ruff check python/turboapi/__init__.py python/setup.py tests/test_release_version_sync.py
• git diff --check
• built turboapi-1.0.32.tar.gz and verified PKG-INFO reports Version: 1.0.32

After merge, tag v1.0.32 to run the publish workflow. The release will not be considered complete until the PyPI ARM64 wheel is downloaded and smoked inside Apple container.

[github-actions]

Performance Regression Report

Runner: ci-pr | Duration: 5s per endpoint | Threads: 4 | Connections: 100 | Workers: 24

Endpoint	req/s	avg latency	p99 latency	threshold	status
GET /health	100,471	0.18ms	0.90ms	80,000	OK
GET /	98,644	0.18ms	1.07ms	80,000	OK
GET /json	95,363	0.21ms	0.94ms	75,000	OK
GET /users/123	95,927	0.20ms	0.99ms	75,000	OK
POST /items	57,294	5.46ms	124.29ms	45,000	OK
GET /status201	91,706	0.19ms	0.74ms	65,000	OK
AVERAGE	89,901			50,000	OK

✅ All endpoints pass regression thresholds