Compare long numeric prerelease identifiers without integer conversion - #488

Fixes #487.

A valid version with a 5,000-digit numeric prerelease parses successfully but raises ValueError when compared under Python’s default integer-conversion limit. Compare ASCII numeric prerelease identifiers by normalized digit count and text, preserving numeric-before-text ordering without converting them to integers.

The change retains constructor leading-zero comparison behavior, legacy Unicode handling and eager validation. It leaves core version parsing, release/build precedence and the optional native parser unchanged. Regression tests and a changelog fragment are included.

Validation on CPython 3.14.7 with the 4,300-digit limit unchanged:

• Baseline: six regression failures; 80 comparison cases passed.
• Candidate tests/: 450 passed, one optional-native-backend skip. After annotation and test-literal/formatting-only cleanup, all 86 comparison cases passed again; the algorithm and test AST were unchanged.
• Ruff found no new lint or formatting findings against the exact baseline. Both whole-file commands remain nonzero for 74 existing diagnostics and 10 existing formatting groups.

The supported-Python tox matrix, native backend, coverage and documentation/doctest builds were not run.

OpenAI Codex assisted with the implementation, tests, independent review and this description.