Motivation

In a multi-journal bookie, CheckpointSourceList builds an aggregate
checkpoint by sampling journals one at a time. The sampling is not an
atomic snapshot, so a periodic SyncThread checkpoint can interleave
with a DbLedgerStorage cache-pressure flush:

1. SyncThread samples journal 0 at position 1.
2. The cache-pressure flush samples journal 0 at position 2 and journal 1
   at position 1, producing [2, 1].
3. SyncThread then samples journal 1 at position 2, producing [1, 2].

The two checkpoints are incomparable because neither covers all state
represented by the other. However, the current lexicographic comparison
treats [2, 1] as later than [1, 2]. DbLedgerStorage can therefore skip
the flush required for [1, 2].

SyncThread can still mark [1, 2] as complete, allowing the journal
checkpoint and cleanup position to advance while entries covered by the
second journal remain only in the write cache. If the bookie crashes
before a later flush, those entries may no longer be replayed during
recovery, resulting in potential data loss.

This requires multiple journals and overlapping checkpoint sampling.
Single-journal checkpoints are not affected.

Changes

• Add a strict coverage comparison for checkpoints while preserving the existing total compareTo ordering.
• Compare aggregate checkpoints component by component.
• Use coverage comparison for the DbLedgerStorage checkpoint skip decision.
• Add unit coverage for aggregate checkpoint ordering and a DbLedgerStorage regression test.

Tests

• mvn -pl bookkeeper-server -Dtest=CheckpointSourceListTest,DbLedgerStorageTest,BookieMultipleJournalsTest,SyncThreadTest,EntryMemTableTest -DfailIfNoTests=false test
• mvn -pl bookkeeper-server -DskipTests checkstyle:check spotless:check apache-rat:check
• mvn -pl bookkeeper-server -DskipTests -Dspotbugs.onlyAnalyze=org.apache.bookkeeper.bookie.CheckpointSource,org.apache.bookkeeper.bookie.CheckpointSourceList,org.apache.bookkeeper.bookie.storage.ldb.SingleDirectoryDbLedgerStorage spotbugs:check