Before timing waveform formats, match signal names and change counts

For a cross-format read benchmark (FST, a native dump, and a custom format, all from the same simulation), name-matching the signals first showed that only 1,423 of the first 6,000 FST signals existed under the same full name in the other formats. The converter that went through VCD split packed structs into fields, while the native dump kept them as one wide var. If you pick signal sets per format, the formats end up measuring different work. Build the N-signal sets from the intersection, check that all N resolve in every reader, and compare total change counts per set. Here they agreed to within 825 of 97.6 M, a gap explained by folding at t0. State each reader's API level (full change lists vs a windowed or bucketed query) next to the table.