[fix][broker] Preserve replicated subscription activity on activation - #26780

Motivation

After #26655, messages published while the replicated subscriptions controller is inactive no longer update the snapshot activity timestamp. If those messages are consumed and their ledgers are trimmed before subscription replication is enabled, both activation seed checks see zero retained entries, leaving snapshots paused until another message is published.

The controller also schedules its first snapshot before initializing its statistics object, allowing the zero-delay task to encounter a null stats reference and leave the pending snapshot gauge incremented.

Modifications

• Remember max-read-position activity for the lifetime of the topic instance and use it when seeding the snapshot timestamp, including after ledger trimming. Topics without a controller continue to avoid reading the wall clock on each publish.
• Initialize controller statistics before scheduling the first snapshot task.
• Add regression coverage in existing test classes for activation and reactivation after trimming, constructor scheduling, and transaction-buffer recovery with replicated subscriptions enabled.

Verifying this change

• Make sure that the change passes the CI checks.

This change added tests and can be verified as follows:

• ReplicatedSubscriptionTest: publish, consume, acknowledge and trim data through the real broker/client path, then enable subscription replication and verify that a cross-cluster snapshot completes without another publish. Covers initial activation, reactivation and keeping empty topics idle.
• ReplicatedSubscriptionConfigTest: execute the first scheduler tick during controller construction using a real topic, marker publication and statistics object.
• TopicTransactionBufferRecoveryTest: publish while the real snapshot processor is paused, then verify that recovery advances the snapshot activity timestamp. Covers recovery with and without a persisted snapshot.

Both bug-fix regressions failed before their corresponding fixes. Local validation passed 22 tests and quickCheck:

./gradlew :pulsar-broker:test \
  --tests 'ReplicatedSubscriptionTest.testEnableReplicatedSubscriptionAfterConsumedDataIsTrimmed' \
  --tests 'ReplicatedSubscriptionTest.testReplicationSnapshotStopWhenNoTraffic' \
  --tests 'ReplicatedSubscriptionsControllerTest' \
  --tests 'ReplicatedSubscriptionConfigTest' \
  --tests 'TopicTransactionBufferRecoveryTest' \
  quickCheck \
  -PtestRetryCount=0 -PtestMaxParallelForks=1 --max-workers=4

Does this pull request potentially affect one of the following parts:

• Dependencies (add or upgrade a dependency)
• The public API
• The schema
• The default values of configurations
• The threading model
• The binary protocol
• The REST endpoints
• The admin CLI options
• The metrics
• Anything that affects deployment

Initializing statistics before scheduling prevents the first snapshot task from failing after incrementing the pending snapshot gauge. Metric names and definitions are unchanged.