Use guarded_eval for the callable in python_func_kw_matches - #15338

Keyword completion is the one evaluation site left in the completer that never consulted the policy:

• python_func_kw_matches resolves the callable with a bare eval(), while global_matches, _evaluate_expr and the dict key matcher all go through guarded_eval
• python_func_kw_matcher is in the default matcher list, so obj.attr( plus Tab runs a property getter or a custom __getattr__ with no cell executed
• guarded_eval refuses that same attribute access under the default limited policy, so the two currently disagree about the same expression

Routed it through the EvaluationContext the sibling matchers already build. Plain functions and ordinary methods still complete their keywords, and the eager behaviour stays available under evaluation="unsafe".