P33 closes RV32IM setup timing

P33 is the timing cleanup pass for P32.

P32 had a credible 32-cycle divider, but multiply and high-half multiply were still wide combinational product paths. The P32 harden run completed physical signoff, but setup timing failed: worst setup slack was -7.798 ns with 474 setup violations.

The fix was to make MUL, MULH, MULHSU, and MULHU iterative. The core now enters S_MUL, runs a 32-cycle shift/add multiply, then writes back either the low or high product word. Divide/remainder stay on the 32-cycle restoring divider from P32.

Commands:

make -C projects/33_rv32im_timing_fix/test clean all
make -C projects/33_rv32im_timing_fix/test act4
make -C projects/33_rv32im_timing_fix/test act4_m
make harden PROJECT=33_rv32im_timing_fix

Results:

Final artifacts:

There were two harden runs. The first P33 run proved the RTL change fixed setup: setup violations dropped to 0, but the default 0.75 ns transition constraint left 5824 max-slew and 88 max-cap violations. The second run used the sky130-realistic MAX_TRANSITION_CONSTRAINT: 1.5 and stronger repair margins. That improved max slew to 2241, but max cap rose to 291.

The practical result: the architectural timing problem is fixed, but electrical DRV cleanup remains. The next useful pass should look at reset/load-mode fanout and backend repair choices instead of adding more ISA features.