[kinqsradio]

Fixes #160307.

Final incremental decoding could return with input still buffered when an error handler consumed only part of an incomplete sequence. For example, decoding b'\x8f\x00' as EUC-JP with surrogateescape and final=True returned '\udc8f' instead of '\udc8f\x00'. The same issue affected EUC-JIS variants and GB18030, including reads through TextIOWrapper.

Resume ordinary decoding after each final error-handler call, repeating final error handling if another incomplete fragment remains. Add regression coverage for chunk boundaries, partial custom error handlers, exception recovery, and file reads, plus a NEWS entry.

Validation on a macOS ARM64 debug build:

• Regression tests produced 21 failing subcases before the fix and pass afterward.
• Relevant codec and I/O suites pass: 1,812 tests run, 44 skipped.
• ./python.exe -m test test_multibytecodec -R 3:3 passes with no leaks in measured repetitions.
• 924,296 incremental-versus-stateless comparisons pass, covering every one- and two-byte input for the four affected codecs with surrogateescape, plus deterministic longer inputs with four error handlers.
• make patchcheck and git diff --check pass.

AI assistance was used to investigate the issue, prepare the patch and tests, and run local validation.

• Issue: euc_jp and gb18030 incremental decoders drop input after an incomplete trailing sequence under surrogateescape #160307