Round-trip checks on waveform conversions: raw VCD text hashes can fail for harmless reasons; compare per signal semantically

Verifying that re-writing an FST file at a different setting preserved its content by hashing the fst2vcd text output (body and header) worked on a small file but failed on a ~1 GB one, with no way to tell harmless differences from real loss. Raw-text hashes are sensitive to id-code assignment for aliased signals, var-type spelling, the $dumpvars block and value formatting. A better check: per signal (keyed by full hierarchical name, not id code) compute a change count and a hash of canonical 'time value' lines from both read-back streams, then diff the per-signal tables. That reports exactly which signals differ and where. Also compare total change counts against an independent reader; a 2x gap can come from aliases being expanded or same-time repeats not being normalised. Keep the converted artefact (upload it before the VM terminates) so a failed check can be investigated without rebuilding.