Reject non-ASCII digits and trailing newlines in version parsing - #478

Version.parse() currently accepts characters outside the SemVer grammar. For example, Version.is_valid('1.2.3\n') is true, and Version.parse('1\u0662.2.3') silently converts a mixed ASCII/Arabic-Indic major component to 12.2.3. Non-ASCII digits can also survive in prerelease identifiers.

The shared regex uses Python's Unicode-aware \d and $, which may match before a final newline. Compile both strict and optional-component patterns with re.ASCII and use \Z to require the actual end of the input. These constraints follow the ASCII digit grammar and prerelease requirements in https://semver.org/#backusnaur-form-grammar-for-valid-semver-versions.

This intentionally rejects previously accepted malformed input; callers wanting whitespace tolerance can strip it before parsing. No exploit or CVE is claimed.

Validation, Windows / Python 3.12:

• All 23 new regression cases fail on the original code.
• Without the optional native backend: 385 passed, 1 skipped; 99.1% coverage.
• With fast-semver-rs-backend 0.1.2: 386 passed; 99.7% coverage, including the fallback after native rejection.
• Ruff passes on the new test file. The changed implementation file has the same 43 existing Ruff diagnostics as base 0f18aa0; mypy has the same existing missing-backend-stub diagnostic on both revisions.
• The two documentation symlinks under tests/ were temporarily materialized as copies for Windows testing, then restored. They are not part of the diff.

AI assistance: Codex helped implement and test this change. The reproduction and verification results are supplied for maintainer review.