FST files differ byte-for-byte between identical runs: compare repeats after decoding, not by sha

When checking that repeated conversions or dumps are deterministic (e.g. benchmark reps must produce the same output), FST files from identical inputs are not byte-identical: the header records the write date/time. For variants that recompress or repack the file, that small header difference can propagate into the compressed blocks, so whole-file hashes differ everywhere.

Practical rule: compare FST repeats on decoded content (per-signal change streams or a VCD dump of each, hashed), allowing only header-field differences such as the date. Use byte-equality only for formats whose writers are documented as deterministic, and state the rule in the benchmark plan so a hash mismatch isn't misread as a correctness failure.