In case CONFIG_HIGH_RES_TIMERS is not set (or when it does not work
as expected), the lamport clock could "stop" in some extremely
rare cases. Theoretically, a symlink update could be missed, or
two transaction log records could accidentally get the same
timestamp. We want any timestamps to be unique (at least on
the same host).
This patch ensures that true forward stepping always takes place,
even when the system clock fails (or at other problems).
For now, the dependency from CONFIG_HIGH_RES_TIMERS is left in Kconfig
as a precondition for MARS.
After some tests and some observational time, it could probably be removed
some day.