C API: Fix PyUnicodeWriter_WriteUTF8() and PyUnicodeWriter_Format() error handling

[vstinner]

If PyUnicodeWriter_WriteUTF8() or PyUnicodeWriter_Format() fails, the writer should remain valid and usable for further writes. These functions can change the buffer kind to support write larger character. But if these functions fail, PyUnicodeWriter_Finish() may need to change the buffer kind to use the most efficient storage for the final string.

Linked PRs

• gh-158439: Add _PyUnicodeWriter.recheck_maxchar #158442

[vstinner]

The fix in the main branch changes the _PyUnicodeWriter structure (to add a new recheck_maxchar member). The structure is private but exposed in the public C API. I prefer to leave the code as it is in 3.14 and 3.15 branches: it's too risky to change the ABI. So I close the issue.