The repository clock started fresh at Hlc.zero on every launch and was
never re-seeded from the database. After a restart, if the device wall
clock lagged behind stamps already stored (e.g. rows imported from a
peer whose clock ran ahead), a local edit was stamped *older* than the
row it edited and silently lost last-writer-wins — so lot changes made
on one device did not update when the backup was imported on another.
Re-seed the clock lazily from the newest stored `updatedAt` before the
first local stamp of the session, mirroring what importInventory already
does with the incoming max. _stamp() is now async.
Tests: add a lot-level LWW re-import test (the variety case was covered,
the lot case was not) and a restart-regression test that reproduces the
lagging-wall-clock scenario.
vjrj
2ea8d67cb4