Implement iterative queue-based traversal in wreduce pass to propagate
width reductions across dependent cells and wires. Previously, wreduce
would process all cells once, then all wires once. This meant that
reductions couldn't propagate through chains of operations.
The new algorithm maintains work queues for both cells and wires,
processing them iteratively until no more reductions are possible.
When a cell or wire is reduced, dependent cells and wires are added
back to the queues for reprocessing.
Add regression test to verify that width reductions propagate through
a chain of operations: (a + b)[3:0] + c, ensuring the first addition
is reduced from 9 bits to 4 bits.