[python-package] Exclude weight_column from text Dataset features

[shiyu1994]

Summary

• fix text-file Dataset loading so metadata columns (including weight_column) are excluded from model features
• keep metadata extraction for weight/group columns unchanged
• add a regression test covering file-based training + prediction input shape

Testing

• PYTHONPATH=python-package /root/miniconda3/envs/lightgbm/bin/python -m pytest tests/python_package_test/test_engine.py -k "text_dataset_with_weight_column_excludes_weight_from_features" -q
• PYTHONPATH=python-package /root/miniconda3/envs/lightgbm/bin/python -m pytest tests/python_package_test/test_engine.py -k "dataset_update_params" -q
• PYTHONPATH=python-package /root/miniconda3/envs/lightgbm/bin/python -m pytest tests/python_package_test/test_engine.py -k "feature_name" -q

Fixes #6856

[jameslamb]

This is not passing the tests. I'll mark it in progress for now, @ me for a review when it's ready and I'd be happy to look.