[Afloat16]

Direct GridSearchCV over STLSQ calls set_params(...).fit(...), so returning None currently breaks the estimator protocol. This patch returns self from BaseOptimizer.set_params and TrappingSR3.set_params after their existing parameter-update behavior.

Six regressions cover empty and changed updates, direct and nested grid searches, and independently derived cross-validation scores. On the current upstream baseline c4421fcec275c8f4cc5c1e93bebb961b212067ae, the new cases give 5 failures and 1 passing control. With the patch, all 6 pass.

Validation on the three changed files:

• Black 23.12.1: formatting unchanged.
• Flake8 7.3.0 with the repository .flake8: no findings.
• Reorder Python Imports 2.5.0: no changes.

Focused tests ran on CPython 3.12, NumPy 2.3.5, SciPy 1.17.0, scikit-learn 1.8.0, cvxpy 1.9.3 and pytest 7.4.4:

python -m pytest --noconftest test/test_optimizers/test_optimizer_set_params.py -q

The shared upstream conftest.py imports JAX, which was unavailable in this environment; these tests use no shared fixtures. The full optional test suite and documentation build were not run. Upstream CI is pending.