fix: reject leading-zero octets in ipv6 IPv4-mapped addresses

[NishchayMahor]

Fixes #1911.

Same class of bug as #1904, but on the ipv6 side. The two IPv4-mapped branches of types.ipv6 (::ffff:a.b.c.d and x:x::a.b.c.d) embed the octet subpattern 25[0-5]|2[0-4][0-9]|1?[0-9]{1,2}, which accepts 2-digit leading-zero octets like 01 and 00. So ::ffff:01.1.1.1 matched even though the stdlib rejects it:

>>> ipaddress.ip_address("::ffff:01.1.1.1")
ValueError: '::ffff:01.1.1.1' does not appear to be an IPv4 or IPv6 address

Swapped the embedded octet for the canonical no-leading-zero form (25[0-5]|2[0-4][0-9]|1[0-9][0-9]|[1-9]?[0-9]) in both branches. Left the standalone ipv4 type to #1904 so these stay independent.

Added the leading-zero cases to the parametrized test. pytest tests/types/test_custom_types.py gives 125 passed.

Thanks @ErenAta16 for filing this with a clean repro.

[github-actions]

📚 Documentation preview: https://dottxt-ai.github.io/outlines/pr-preview/pr-1936/

Preview updates automatically with each commit.

[RobinPicard]

Looks good, thanks!

[ErenAta16]

Verified the regex change, [1-9]?[0-9]|1[0-9][0-9] correctly excludes the leading-zero forms while still accepting a bare single digit. Good test coverage on both the rejected and accepted cases. Thanks for picking this up.