feat(api): cap request output tokens

[sufubao]

Summary

• add --max-request-output-tokens as the default and hard output-token limit
• apply the cap centrally across OpenAI, TGI, and native APIs
• document the option in English and Chinese

Tests

• PYTHONPATH=. pytest -q test/test_api/test_max_request_output_tokens.py test/test_api/test_seed_validation.py
• CLI parsing for hyphenated and underscored option names
• git diff --check