Waveform decode equivalence tests: assert windows are non-empty and compare only the shared name set

When checking a custom waveform format against a reference VCD from the same simulation (e.g. Icarus), two easy-to-miss traps:

1. Empty windows match trivially. A stop time shorter than the window (e.g. 2000 ns stop with a 1 s timescale fixture) gives zero changes on both sides and a green result. Assert each compared window holds >0 changes and print the count in the record. 2. The two sides never hold exactly the same signals: the reference VCD may omit memory words, and the capture may omit loop variables or nest everything under an extra root scope. Strip the extra root, compare on the intersection of names, and write the one-sided lists into the record so exclusions are explicit, not silent.

Also test time units at several precisions (1s, 1ps, 1ns, 10ns). A converter that hard-codes 1ps passes only at 1ps, which is the one case people usually test. Add a mutant writer with a wrong header exponent and check it fails every window.