[improve][broker] Skip unused replicated subscription timestamp updates - #26655

Motivation

Every successful normal publish advances the transaction buffer's maximum readable position. That path currently reads the wall clock and writes lastMaxReadPositionMovedForwardTimestamp, even when replicated subscriptions are not enabled for the topic.

ReplicatedSubscriptionsController is the only production consumer of this timestamp. On ordinary topics, the clock read and volatile write therefore add work to every publish without affecting broker behavior.

Modifications

• Skip the timestamp update when the topic has no replicated-subscription controller.
• Seed the timestamp for topics with stored entries immediately before controller construction and again after publishing the controller reference. Empty topics remain idle, while the initial snapshot covers earlier messages and publishes racing with controller activation.
• Preserve marker handling, maximum-readable-position updates, transaction recovery, and snapshot scheduling behavior.
• Add lifecycle, transactional, recovery, and callback microbenchmark coverage, including empty-topic activation and a publish during controller construction.

Performance

An actual-callback JMH benchmark measured the controller-absent path:

Allocation remains effectively zero. When a replicated-subscription controller is active, the added presence check measured 20.090 ns/op versus 19.360 ns/op for the unconditional baseline (+0.730 ns/op).

A five-million-message broker scenario with 500 producers and five Key_Shared subscriptions produced two positive matched comparisons:

The spread indicates substantial whole-host variance, so these runs establish repeatability and absence of a throughput regression rather than a precise effect size. Every run delivered all messages to every subscription with zero duplicates, ordering violations, or invalid messages.

Verifying this change

This change added tests and can be verified as follows:

• Reproduced ReplicatedSubscriptionTest.testReplicationSnapshotStopWhenNoTraffic and the empty-topic lifecycle regression on the original code.
• After the fix, all 15 tests in ReplicatedSubscriptionTest and ReplicatedSubscriptionsControllerTest passed with -PtestRetryCount=0.
• Focused persistent-topic and replicated-subscription controller lifecycle tests
• Transaction commit and transaction-buffer recovery tests
• ./gradlew :microbench:shadowJar quickCheck
• JMH with two forks, three 1-second warmups, five 1-second measurements, and GC profiling

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