Unbounded-Time Safety Verification of Guarded LTI Models with Inputs by Abstract Acceleration