Match binascii hex separator validation - #9643

Summary

Align binascii.hexlify and b2a_hex separator validation with CPython's bytes-output path. Check the object's length first, then accept a str containing only Latin-1 code points or a bytes object, including subclasses. This permits separators such as b'\xff' and 'é' while rejecting length-one bytearray and memoryview separators. It also preserves the observable __len__ callback and reads the first underlying character or byte when a subclass overrides its reported length.

ASCII strings use the cached kind fast path. For a non-ASCII string subclass that reports length one despite having longer underlying content, checking whether all code points fit in Latin-1 scans the string; CPython caches its internal string width. Generic argument-binding diagnostics, lower-level bytes-per-separator conversion, output-buffer lifetime, and allocation limits are outside this change.

Validation

• Compared CPython 3.15.0 binascii and pystrhex source and reproduced high-byte/Latin-1 separators, out-of-range characters, buffer rejection, callback ordering and subclasses with overridden __len__.
• Independent source review, normal commit hooks and diff checks passed.
• RustPython was not built or executed locally. No tests were added or changed; hosted CI can run existing test_binascii coverage.