Fixes the broken `_total.h5` virtual-dataset (VDS) pointers described in the
companion discussion. For `ID19`-`ID68` (50/68 participants), the `_total.h5`
virtual sources point at the wrong participant's `_part_N.h5` files (all off
by 18 folders, e.g. `ID19_total.h5` -> `ID01_part_1.h5`). Since those files
don't exist in the affected participant's folder, HDF5 silently fills the
unresolved virtual sources with zeros instead of raising, so reading through
the documented `_total.h5` path currently yields an all-zero recording for
those 50 participants.
This PR replaces all 68 `_total.h5` files with regenerated versions that
point at the correct `_part_N.h5` files (read from each file's own
`info/files`, which already has the right names -- only the VDS pointers
were wrong). No signal bytes are touched; `data/seizures`, the channel/
sampling-rate attributes, and `info/files`/`info/checksums` are carried over
unchanged, and the dataset's total shape is unaffected.
Verified: after this ch…
