ktrdump: ignore ts=0 when searching for earliest_ts()
When merge-printing multiple cpu buffers, we already treat ts=0 as
a condition to prefer a more recent entry. However when searching for
the first entry, ts=0 (empty) will be treated regularly. This can lead
to a situation that ktrdump would only print entries from the last CPU:
Assume you had 4 CPUs, and the buffer for CPU #2 and #3 started out with
empty entries (which would not be ignored by earliest_ts()). When
searching for the next entry, the empty (ts=0) entry of CPU #2 would
always be selected as the first entry. However a ts=0 entry of CPU #3
would override this. In this case only the index of CPU #3 would
advance until full entries would be printed. Once in this situation,
processing the ts of CPU #2 would always reset ts to 0, and this would
be treated as "not found" when processing CPU #3's entries, leading to
an output that only contains CPU #3 entries.