test: read generated outputs as UTF-8 on Windows

[xumian520]

Summary

• read generated graph JSON and HTML using their actual UTF-8 encoding
• keep merge-chunks Unicode IDs intact under non-UTF-8 Windows locales
• avoid changing production output or global locale behavior

Reproduction

On a Chinese Windows environment, the affected tests decoded UTF-8 output with the default GBK codec. Pipeline tests raised UnicodeDecodeError, and the merge-chunks Unicode ID assertion observed mojibake.

Verification

• pytest -q tests/test_merge_chunks_validation.py tests/test_pipeline.py — 18 passed
• affected files plus optional-backend environment checks — 22 passed
• ruff check tests/test_merge_chunks_validation.py tests/test_pipeline.py — passed
• git diff --check — passed

The repository has additional pre-existing Windows-specific test failures outside these files; they are intentionally not included in this focused change.