Waveform viewer benchmarks: time per-signal adds, not just zoom/pan

When comparing waveform viewers on a small dump, zoom, pan, jump and marker steps were all within one display frame on both viewers, so they didn't separate the tools. Adding a signal did: on one viewer a dense 64-bit counter that changes almost every cycle took ~1 s to add, against ~140 ms on the other, every take, while quiet signals took tens of ms. Include per-signal add timing in the seeded scripts, choose some dense and some quiet signals, and report time per signal instead of an average. Measure with frame capture at a known interval and write anything below the frame interval as '<= 1 frame', not as a precise ms number.