Fix hyperlink target detection in URL wrapping

[jum-apzn]

Fixes #269.

The malformed prefix alternative in URL_REGEX excludes explicit reST hyperlink targets. A long target can consequently be split away from its URL, producing invalid markup.

This separates the inline-link and explicit-target alternatives, restores periods in inline labels, and keeps the target underscore mandatory. The angle-bracket lookahead prevents an earlier reference from being consumed with a later anonymous URL, preserving the behavior covered by #75 and #210.

Adds 18 parametrized cases covering complete link spans, non-link text, unrelated markup, long targets at module/function indentation, and formatting idempotence. Seven of those cases fail before the fix.

Validation on Python 3.12.14:

• Full pytest suite: 621 passed, 3 skipped
• 24 additional CLI preservation/idempotence checks
• Black, Ruff, isort, mypy, pydocstyle, docformatter, and git diff --check passed

Related prior work: #324 included a similar regex correction in a broader change; this keeps the fix focused on #269 with regression coverage.