Benchmarking waveform dump formats from a live simulator: compare total wall, not just dump overhead, and run each format at its smallest setting

When comparing trace formats captured live from a simulator (e.g. FST vs a custom format vs a vendor format), report three numbers per mode: total wall/CPU, overhead vs a no-dump run, and file bytes. On a small trace (47 signals, 200K cycles), a 30% cut in dumping overhead was only a 0.8% cut in total wall time, because simulation dominates. Run each competitor at its smallest documented setting. FST's fstWriterSetRepackOnClose(1) shrank LZ4 output from 464 KB to 106 KB without changing events, and on this live trace LZ4+repack came out smaller than zlib+repack. That differs from runs that convert from VCD, so check per corpus instead of assuming. Also make sure your own format is at its best level before quoting a size gap. Check every output event for event against a reference reader. Record a CPU audit of other processes before each rep, and mark reps CONTENDED when background load is high. Readers return data in different layouts, so call full-read timings implementation comparisons, not format-intrinsic ones.