Phase 1: select a self-contained static liburing backend

[bright2227]

Summary

This PR completes the selected Phase 1 native-ring backend work:

• add the static-liburing native Ring lifecycle implementation
• record the comparison measurements and select pinned, statically linked liburing
• package the minimal pinned liburing sources in source distributions
• build a private position-independent liburing.a during extension builds
• link both the existing CFFI backend and native liburing implementation without a runtime liburing.so dependency

The raw-syscall experiment is intentionally not part of this PR. It is preserved separately on feature/raw-syscall-ring-spike for future reference.

Decision

Use the pinned static-liburing route for the production native core. The lifecycle syscall counts were identical and local timing was too noisy to distinguish the compared implementations. The modest binary-size increase is preferable to maintaining correctness-sensitive SQ/CQ shared-memory ordering and kernel ABI details directly.

The vendored build keeps the selected liburing version consistent across Linux distributions. Runtime io_uring feature availability still depends on the host kernel.

Scope

The existing CFFI proactor remains the active backend in this PR. The native liburing module establishes the lifecycle boundary and backend decision; migration into the canonical production module remains follow-up work.

Validation

• .venv/bin/pytest — 25 passed
• Ruff and Markdown formatting checks pass
• clean source distribution builds into a wheel
• the clean wheel contains _uringcore_liburing and no raw _uringcore extension
• both packaged liburing extensions link the private archive without liburing.so
• final automated code review found no actionable issues