Version.CheckOrdering firing for L0 tables with zeroed seqnums

[petermattis]

The metamorphic test I'm working on is bearing fruit:

=== RUN   TestMeta
...
// INFO: [JOB 32] flushing to L0
// INFO: [JOB 32] flushing: sstable created 000031
// INFO: [JOB 31] compacting: sstable created 000032
// INFO: write stall beginning: memtable count limit reached
// INFO: [JOB 32] flushed to L0: 1 (849 B)
// INFO: [JOB 33] WAL created 000033 (recycled 000026)
// INFO: write stall ending
db.DeleteRange("Nj7E5S.s3", "UErmWaZ") // <nil>
// INFO: [JOB 31] compacted L0 -> L6: 3+0 (2.4 K + 0 B) -> 1 (833 B)
// INFO: [JOB 34] flushing to L0
// INFO: [JOB 34] flushing: sstable created 000034
db.Get("sLpYRDDCLr") // ["VODVmPX9JAL8t7e"] <nil>
// INFO: write stall beginning: memtable count limit reached
// INFO: [JOB 34] flushed to L0: 1 (888 B)
// INFO: [JOB 34] sstable deleted 000019
// INFO: [JOB 34] sstable deleted 000025
// INFO: [JOB 34] sstable deleted 000027
// INFO: [JOB 34] sstable deleted 000034
// INFO: background error: pebble: internal error: L0 files 000031 and 000034 are not properly ordered: 13-13 vs 0-0
exit status 1
FAIL	github.com/cockroachdb/pebble/internal/metamorphic	0.024s

In this test run, the memtable size was set to 1000 which creates lots of tiny L0 sstables. I believe 000031 had keys which overlapped with a lower-level, but 000034 did not.

[petermattis]

From another run:

// INFO: [JOB 7] flushing: sstable created 000008
// INFO: [JOB 7] flushed to L0: 1 (887 B)
// INFO: [JOB 7] sstable deleted 000008
// INFO: background error: pebble: internal error: L0 file 000008 does not have strictly increasing largest seqnum: 0-0 vs 0