nir/opt_if: Don't split ALU if it's already at the bottom of the loop
If the ALU instruction is already at the bottom of the loop, there's no need to perform the split.
This prevents an infinite loop (applying this optimization and growing the shader again and again) in dEQP-VK.reconvergence.subgroup_uniform_control_flow_ballot.compute.nesting2.3.2 and various others reconvergence tests.