Expose load_ml_source on the ML load stats entity

[Speshman]

This is my first contribution to Predbat, so feedback on repository conventions is especially welcome.

Fixes #4907

Summary

• Expose the configured load_ml_source value on sensor.predbat_load_ml_stats.
• Add a regression assertion covering the False value, distinguishing observation-only use from ML driving the plan.
• Document the new attribute alongside the existing Load ML statistics.

Testing

• ./run_all --test load_ml: 34/34 passed.
• The focused publishing regression fails without the implementation (KeyError: 'load_ml_source') and passes with it.
• Ruff, Black, Markdownlint, CSpell, whitespace, YAML/JSON, and repository-specific pre-commit hooks passed.
• GitNexus impact and change analysis reported low risk.

Notes

The complete ./run_pre_commit wrapper reaches the quick suite but does not finish in this environment because of an existing order-dependent deadlock in test_db_manager_error_handling after earlier test groups. That test passes in 1.5 seconds alone and also passes with its adjacent download and axle groups. The changed load_ml suite and all static checks pass.

Prepared with assistance from GPT-5.6 Sol.

[chalfontchubby]

Thanks @Speshman .