Pin utf-8 in add-new-model-like file IO (Windows non-UTF-8 locale fix)

[dltsum]

What

transformers add-new-model-like reads and writes repository files with the locale's preferred encoding. On Windows with a non-UTF-8 locale (e.g. cp936) the command crashes when writing the generated doc file, whose header contains ⚠️:

src/transformers/cli/add_new_model_like.py:599: in add_new_model_like
    f.write(doc_file)
E   UnicodeEncodeError: 'gbk' codec can't encode character '\u26a0' in position 604: illegal multibyte sequence

(found by tests/utils/test_add_new_model_like.py::TestAddNewModelLike::test_llama_without_tokenizers on a cp936 machine)

Fix

Pin encoding="utf-8" on all seven open() calls in the command — everything it reads (_toctree.yml, existing test files) and writes (modular file, __init__.py, test files, doc file) is UTF-8 by repository convention.

Verification

Windows 11, cp936 locale, Python 3.14:

• The doc-file content fails to write with a bare open(path, "w") (UnicodeEncodeError: 'gbk' ... '\u26a0') and succeeds with encoding="utf-8".
• With the fix, test_llama_without_tokenizers progresses past the previous crash point. Note: the test still fails later in this environment for unrelated reasons — its subprocesses resolve the installed site-packages transformers (whose tokenizers<=0.23.0 pin conflicts with the newer tokenizers needed by main), and the temp-repo check_doc_toc.py step errors on the partial doc copy. Both are local-environment artifacts, not affected by this change either way.

🤖 Generated with Claude Code

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 35441048549:2
Result: success | Jobs: 2 | Tests: 20 | Failures: 0 | Duration: 2m 20s