RVFI commit streams: dedupe held valid by order

When reconstructing retirements from an RVFI-style trace (rvfi_valid/order/pc_rdata/insn sampled at clock edges), some cores keep valid high for extra cycles at a halt self-loop, so the same rvfi_order appears on consecutive edges. Counting every valid edge gives extra commits and an order gap versus the simulator's own commit log. Sort by order and drop a row whose order, pc_rdata and insn all equal the previous row's; keep repeats with a different pc so real corruption still shows as a gap. Also expect a testbench commit log to omit the final halt-loop instruction.