Add optional native parser backend (minimal adapter) - #1

Fixes python-semver#474

Summary

This PR implements the minimal adapter for an optional native parser backend as proposed in python-semver#474. The implementation is intentionally small to minimize maintenance footprint:

• 3 files changed, 41 lines added
• Pure Python remains the default — no changes to existing behavior
• Native backend is opt-in via pip install python-semver[native]
• Backend package (fast-semver-rs-backend) is maintained separately by the contributor
• Automatic fallback to Python parser when backend is unavailable or encounters bounded integer limits

Changes

1. pyproject.toml: Add [project.optional-dependencies] with native extra pointing to fast-semver-rs-backend>=0.1.1 (CPython only)
2. src/semver/version.py: Optional import of native backend; use it in Version.parse() only when:
   • Backend is present
   • Not using optional_minor_and_patch mode (Python-only feature)
   • Backend doesn't raise ValueError (falls through to Python for unbounded integers)
3. tests/test_optional_native_backend.py: Tests for backend presence and fallback behavior

Compatibility

• ✅ All 41 existing parsing tests pass
• ✅ Backend test skips gracefully when not installed
• ✅ No breaking changes to public API
• ✅ PyPy and other implementations continue using pure Python path

Maintenance model

• python-semver maintains only the thin adapter (this PR)
• Backend wheels, implementation, and CI are maintained separately by the contributor
• No packaging overhead for core maintainers
• Decoupled release cycles

Verification

Per python-semver#474:

• Official test suite: 342 passed with backend, 341 passed + 1 skipped without
• 10,000-case differential corpus: 0 mismatches
• Backend CI: 10/10 jobs passed across Linux, macOS, Windows (x86_64/arm64) on CPython 3.10–3.14

Reviewer guidance

@tomschr: This PR focuses exclusively on the maintenance footprint of the adapter itself. The backend implementation, wheels, and benchmarks are provided as context in python-semver#474 but are not part of this review scope. Please evaluate:

1. Whether the 41-line adapter is acceptable maintenance overhead
2. The optional-dependency model and fallback behavior
3. Any concerns about the integration points

Performance evidence and detailed backend verification are documented in the original issue.