Bug 1: Size was too small which led to some digis with TS_MSB above 65555 escaping the duplicates filter.
No effect on the digi themselves (time is computed from the original 64b/32b value).
In the legacy case we can stay with 32b (instead of the original 64b or the old 16b) as the field in the raw message is 29b, while in the current version we need 64b to account for potential TS_MSB cycles and large Timeslice start offset.
Bug 2: In both Legacy and current unpackers, a signed integer was used at some point to store/return the raw timestamp in clock cycles. While the last bit would probably never be reached in standard usage, it could lead to negative values and crazy results during the conversion through double (full precision time in ns) and back to unsigned integer (ns precision time) for data coming from badly synchronized setups
Bug 3: In the current unpacker, the subtraction of the Timeslice start time from the TS_MSB value was not protected against the case of badly synchronized data where the TS_MSb does not match the Timeslice index. It now triggers a Fatal as these data are not recoverable with this unpacker.