[vinsymoyya]

Issue #, if available: Fixes #955

Description of changes

Select an explicit manylinux target from the Lambda Python runtime and architecture instead of using a generic GNU/Linux triple. Python 3.12+ uses manylinux_2_34 for AL2023; Python 3.8-3.11 uses the AL2-compatible manylinux_2_17 download target already used by the pip workflow. This prevents choosing wheels that require a newer glibc and allows AL2023 wheels that the generic target rejects on Windows.

UV does not expose an exact manylinux_2_26 target, so AL2 targeting is intentionally conservative: wheels tagged only for glibc 2.18-2.26 are outside the chosen target. Calls without a Python version retain their existing behavior.

This change is independent of #953's requirements paths and #954's lock handling.

Description of how you validated changes

• On unchanged upstream, all 12 new regression cases fail; after the fix, all 71 focused unit, functional, and integration checks pass locally.
• Real offline integration cases offer a compatible wheel and an incompatible newer-glibc wheel of the same version. They verify wheel selection for both architectures, the Python 3.11/3.12 boundary, and both experimental-flag settings. Fixtures use platform-tagged wheel metadata; no native extension is compiled or executed.
• Fork validation for commit 36a0675 passed Linux and Windows with uv 0.11.26 and 0.13.7. Each job passed 1,025 unit/functional tests, 28 UV integration tests, Ruff, Black, and the 94% coverage gate.
• On each host/version combination, the eight real wheel-selection regressions were first run against unchanged upstream and all failed; the same cases pass with the fix.

Checklist

• Review the generative AI contribution guidelines

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.