fix(officecli): use safe Windows bulk writes

[parkjs101]

Summary

• make explicit-cell officecli batch --input the default Windows DataFrame→XLSX path
• add a split-JSON→batch converter that preserves CJK, embedded newlines, and numeric types
• require officecli close plus exact persisted-cell read-back; demote officecli import to verified non-Windows legacy use
• add regression fixtures and unit/real-binary tests

Verification

• OFFICECLI_BIN=C:\Users\user\AppData\Local\OfficeCli\officecli.exe uv run --no-project --with pytest pytest tests/test_officecli_data_pipeline.py -q → 7 passed
• OfficeCLI 1.0.45 legacy import reproduced: command reports 3×3 imported, exit 0, validate passes, but persisted A1 reads (empty)
• safe batch path: 9/9 commands succeeded; closed-workbook read-back matched Korean header/text, embedded newline, and numeric cell; validate passed

Addresses lidge-jun/cli-jaw#279 and lidge-jun/cli-jaw#295.