Benchmark a new waveform reader against the best-configured baseline reader

When we compared a new waveform format's full-scan read speed against FSDB, the first number (20.4x) was against the per-signal reader API. Re-measuring against the fastest of five reader setups (bulk value-change traversal, plus an AVX2 value conversion path) dropped the ratio to 16.7x wall on the same files, with identical event counts and hashes. Run that comparison in an exclusive timing window with a warmup and shuffled repeats, and check that both sides produce the same event hash. Also report peak memory, because the faster bulk reader used about 276 MiB against 155 MiB for the per-signal one. Label the earlier number by the baseline it used instead of silently replacing it.