Avoid redundant state writes on coordinator updates

[MTrab]

Summary

• avoid calling async_write_ha_state() for sensors, binary sensors, and switches when coordinator data does not change entity state
• keep entity attributes untouched when the computed state is unchanged
• add targeted regression tests covering unchanged and changed update paths

Test strategy

• ruff check .
• python -m pytest -q tests

Known limitations

• this only removes redundant writes for the entity types covered here; other platforms are unchanged
• no hardware validation was performed; coverage is based on unit tests only

Configuration changes

• none