Bug in get_act_layer_op

[RobinGeens]

When get_act_layer_op is called for a layer with two constant operands (which is the case for the very first layer), then the activation layer operand is found based on the partially related loops. However, for matmul layers, there are no PR loops and the code fails. In this case, both operands are activation operands.

zigzag/zigzag/workload/layer_node.py

Line 207 in cdfa78e

pr_loop_key = next(iter(self.pr_loop.keys()))