Increment long numeric prerelease and build identifiers without integer conversion - #496

Summary

Version.parse() accepts a valid 5000-digit numeric prerelease or build
identifier, but bump_prerelease() and bump_build() raise ValueError under
Python's default integer-string conversion limit. Increment ASCII decimal
strings directly rather than converting them to integers.

Preserve existing constructor normalization of prerelease leading zeros,
build padding, Unicode digit handling and input immutability. The process-wide
integer conversion limit is left unchanged.

Validation

• Actual base regressions: 5 expected failures and 6 passing controls.
• Full Windows Python 3.11.9 suite: 482 passed, 1 optional-native-backend skip;
  100% statement and branch coverage.
• Independent comparison: 20,412 existing-behavior cases match the base.
• New regression module passes Ruff lint and formatting. Complete repository
  Ruff results remain at the same 160 inherited findings; no added occurrences.
  Upstream marks that check nonblocking. Configured mypy still reports the
  optional backend's missing typing information; docformatter still reports
  existing formatting differences. Neither is claimed to pass.
• The full supported-interpreter, native-backend, documentation and packaging
  matrices have not been run. A numbered newsfragment is deferred until a PR
  number exists, following the project's post-PR changelog instructions.

AI assistance disclosure: OpenAI Codex assisted with investigation,
implementation, regressions and independent baseline verification under the
contributor's authorization.