[core] Support LocalKvDb merge and TTL compaction

[JingsongLi]

Purpose

Add the generic compaction primitives needed to preserve RocksDB-style merge and non-strict TTL semantics in LocalKvDb.

Changes

• Add an optional merge operator for adjacent logical records written by flush and compaction.
• Add an optional expiration predicate evaluated during compaction.
• Convert expired values to tombstones during partial compaction so older values cannot be resurrected.
• Drop expired values during full compaction.
• Apply expiration after record combining, allowing a merged value to refresh its effective TTL.
• Rewrite a single existing run during full compaction when TTL cleanup is configured.

State-specific value envelopes and merge operators are intentionally left to the dependent state-backend PR.

Verification

• mvn -q -pl paimon-common -Pfast-build -DwildcardSuites=none '-Dtest=LocalKvDbTest#testPartialCompactionDoesNotResurrectValueFilteredByExpiration+testCompactionMergesBeforeFilteringExpiredValues+testCompactionRemovesTombstones+testManualCompaction' test
• mvn -q -pl paimon-common -Pfast-build -DwildcardSuites=none -Dtest=LocalKvDbTest test
• mvn -q -pl paimon-common spotless:check -DskipTests
• mvn -q -pl paimon-common -DskipTests compile

[leaves12138]

There is a blocking stale-value resurrection case when a merge operator collapses multiple physical keys while older runs still contain the consumed keys.

[leaves12138]

RecordCombiningWriter can collapse multiple keys into the first key without writing tombstones for the consumed keys. If an older run still contains one of those keys (for example from bulkLoad), reads fall through to the stale value after flush. I reproduced this with old a-1/a-2, then new a-1/a-2: get(a-2) returns new-2 before flush and old-2 after flush. Partial compaction has the same risk when older runs remain. Please ensure every consumed key shadows older versions, or restrict combining to cases where no older run can remain, and add a regression test.