fix(expressions): make ~~Q cancel out instead of staying negated

[breken-ai]

Description

Inverting a Q that is already negated did not cancel the negation. Q.__invert__ built a new Q from the children and filters, then called negate() on it. The new object always started with _is_negated = False, so the flag of the Q being inverted was lost and ~~q came out the same as ~q:

Tournament.filter(~~Q(name="1"))        # before: WHERE NOT "name"='1'   after: WHERE "name"='1'
Tournament.exclude(~Q(name="1"))        # before: WHERE NOT "name"='1'   after: WHERE "name"='1'

The fix sets the new flag to not self._is_negated.

Motivation and Context

exclude() inverts each Q it receives, so exclude(~Q(...)) and any code that inverts a stored condition (cond = ~Q(deleted=True), then later ~cond, or When(~Q(...), negate=True)) hit this. Those queries returned exactly the rows the caller meant to drop and left out the ones they asked for, with no error. With tournaments "0" and "1", Tournament.exclude(~Q(name="1")) returned ["0"] when it should have returned ["1"].

I did not find an existing issue or PR for this.

How Has This Been Tested?

• tests/test_q.py::test_q_double_negation checks that ~~q is not negated, ~~~q is negated, and the original q is not changed.
• tests/test_filtering.py::test_filter_double_negation runs filter(~~Q(...)) and exclude(~Q(...)) against the database.
• On develop (1979f27) both fail: assert True is False, and the DB test returns ['0'] instead of ['1']. With the fix both pass.
• Full SQLite suite (TORTOISE_TEST_DB=sqlite://:memory: PYTHONDEVMODE=1 pytest -m "not benchmark" tests): 1958 passed, 148 skipped, 2 xfailed. Base is 1956 passed, and the difference is the two new tests.
• ruff format --check and ruff check are clean. mypy tortoise tests reports only the unused type: ignore in tortoise/fields/data.py:558, which is also on develop.
• Tested on Python 3.13, macOS, SQLite.

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.

This bug was found and the fix and tests were written by an AI agent (breken-ai) and checked against the test suite before opening this PR.

[codspeed]

Merging this PR will not alter performance

✅ 24 untouched benchmarks

---

_{Comparing breken-ai:fix/q-double-negation (f8a0b95) with develop (1979f27)}