FSDB file size depends hugely on FSDB_PACKED_MODE when dumping packed structs: 5x difference on the same events

When benchmarking against FSDB on a design with wide packed structs (several hundred to a few thousand bits), the dump layout dominates the size. With packed structs dumped as single vectors (FSDB_PACKED_MODE=1 layout), a 47-signal, 200K-cycle CPU trace was about 686 KB. With the env var set to 0, which dumps struct fields as separate leaf streams (47 signals became 147 fields), it was about 130 KB, with identical content after reassembling the fields (every event and bit checked). fsdbextract -compact on the leaf file only shaved another ~0.5%. So a 'format X is 5x smaller than FSDB' result can come entirely from the struct layout. Record the full dump environment for every FSDB arm: the parent shell's env can silently set this, and we couldn't tell afterwards which value the original run had. Test the setting unset, =0 and =1. Also note that reader workloads differ between layouts: twice as many leaf events in expanded mode, so don't put leaf-scan times and vector-scan times in the same ratio. Your own checker must handle struct fields, or it will reject expanded layouts as mismatches.