Clear removed model options in AlterModelOptions

[edu2004eu]

Description

AlterModelOptions.state_forward applied its options with model_state.options.update(...), which can only add or change keys. When an option is removed from a model the generator emits an AlterModelOptions whose options simply don't contain it, so the old value stayed in the state and the same operation was emitted again on the next run.

This clears the keys that belong to this operation and aren't in the new options. The five keys carried by other operations (table, app, indexes, unique_together, constraints) are left alone — that list already existed as a literal in _model_options_for_compare, so I moved it to AlterModelOptions.UNMANAGED_OPTION_KEYS and had the diff use it. The bug was the two sides disagreeing about which keys this operation owns, so it seemed worth having one definition.

Motivation and Context

Fixes #2279.

Deleting a model's docstring clears table_description, and from then on makemigrations writes the same migration on every run and never gets back to No changes detected. In a project with a few such models, every later migration carries a block of no-op operations unrelated to the change being made.

In practice table_description is the only key this affects today, since pk_attr is always present.

Not covered here: database_forward returns None and table comments are only written by CreateModel, so on PostgreSQL a changed or deleted docstring still never reaches COMMENT ON TABLE. Same root cause, but it needs SQL emission per backend, so I left it out of this PR.

How Has This Been Tested?

Three tests, in the files already covering these units:

• test_alter_options_removes_dropped_key — a removed key is gone from the state after the operation applies.
• test_alter_options_keeps_unmanaged_keys — unique_together and table survive, guarding against the clearing being too broad.
• test_removing_table_description_settles_after_one_migration — generates the operation, applies it, and asserts the next diff is empty. This is the reported symptom.

The first and third fail on develop and pass with the change.

Also reproduced end to end before and after: a one-model SQLite project, docstring removed, makemigrations + migrate three times. Before, three identical migrations; after, one, then No changes detected.

make check passes. pytest tests/migrations/ is 302 passed, 6 skipped; the full SQLite suite is 1917 passed, 148 skipped, 2 xfailed. One unrelated failure locally, tests/fields/test_time.py::test_zoneinfo, is a missing tzdata on my machine and fails the same way without this change.

Checklist:

• My code follows the code style of this project.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.
• I have added the changelog accordingly.
• I have read the CONTRIBUTING document.
• I have added tests to cover my changes.
• All new and existing tests passed.

[codspeed-hq]

Merging this PR will not alter performance

✅ 24 untouched benchmarks

---

_{Comparing edu2004eu:fix/altermodeloptions-remove-keys (293d430) with develop (e5bddf2)}