Fix reduction behavior

[igor-sachok]

A reduction instruction whose operand depends on a preceding arithmetic instruction can deadlock: once the reduction enters the Reduction_Reduce stage. The preceding instruction's retire condition in control_proc is never satisfied — it can't retire and release its result. The reduction, in turn, is stuck waiting on that same result. That creates a deadlock.
Code example:
asm volatile("vsetvli %0, %1, e8, m8, ta, ma" : "=r"(vl) : "r"(64));
asm volatile("vmv.v.i v0, 1");
asm volatile("vrsub.vi v16, v0, 0");
asm volatile("vmv.v.i v24, -1");
asm volatile("vredand.vs v8, v16, v24");
asm volatile("vmv.x.s %0, v8" : "=r"(result));
A solution is to add a check to control_proc that non-reduction instructions can retire regardless of reduction_state_q