[improve][broker] Part 2: Pack PendingAcksMap values into long

[void-ptr974]

Motivation

Part of #26027.

PendingAcksMap tracks entries that have been dispatched to Shared and Key_Shared consumers but are not fully acknowledged yet. Before this PR, each pending entry stored its value as an IntIntPair:

• remaining unacked message count
• sticky key hash

The consumer hot paths usually only need the remaining unacked count after a lookup or removal. Keeping a value object per pending entry increases retained heap, and returning IntIntPair from those paths can create short-lived objects that are avoidable.

Modifications

• Pack the pending-ack value into one long while keeping the same outer/inner sorted-map layout and ledger/entry ordering.
  • high 32 bits: remaining unacked count
  • low 32 bits: sticky key hash
• Add primitive accessors for the consumer paths that only need the remaining unacked count:
  • getRemainingUnacked
  • removeAndGetRemainingUnacked
• Update Consumer to use the primitive accessors for ack owner lookup, full ack completion, and redelivery removal.
• Keep the existing get and removeAndGet methods that return IntIntPair for compatibility, and mark the legacy pair-returning path as deprecated.
• Add PendingAcksMapPackingBenchmark under microbench so this value-layout change can be measured independently from later inner-map changes.

This PR only changes the value representation and the hot-path accessors. It does not replace the inner ordered map and does not include the later primitive hash-map or BitSet cleanup-index work.

Correctness coverage

PendingAcksMapTest covers the updated behavior:

• packed value round trip through public accessors
• zero remaining count
• signed sticky key hash values, including negative hashes
• integer boundary values
• value-based removal
• updateRemainingUnacked
• removeAllUpTo callback value preservation
• missing-entry lookups

Benchmark

This PR adds PendingAcksMapPackingBenchmark, which compares the value layout before this PR with the value layout introduced by this PR, using the same ledger/entry ordering behavior in both cases. The table below reports the before/after results relevant to this PR.

Run command:

./gradlew :microbench:shadowJar

java -jar microbench/build/libs/microbench-5.0.0-M1-SNAPSHOT-benchmarks.jar \
  'PendingAcksMapPackingBenchmark\..*' \
  -p entries=50000 \
  -p ledgers=1 \
  -wi 1 -i 3 -w 1s -r 1s -f 1 \
  -bm avgt -tu ns -prof gc

Local OpenJDK 25.0.2 results:

Operation	Before this PR	This PR	Change
addOrReplace	183.9 ns/op, 24.7 B/op	177.9 ns/op, 0.5 B/op	-3.3% time, -98.0% alloc
containsHit	79.3 ns/op, 0.3 B/op	70.6 ns/op, 0.2 B/op	-11.0% time
getRemainingUnackedHit	81.1 ns/op, 0.3 B/op	70.4 ns/op, 0.2 B/op	-13.3% time
updateRemainingUnacked	181.4 ns/op, 24.7 B/op	173.0 ns/op, 0.5 B/op	-4.6% time, -98.0% alloc
removeAndGetRemainingAndAdd	338.3 ns/op, 65.3 B/op	294.9 ns/op, 40.8 B/op	-12.8% time, -37.5% alloc
removeWithValueAndAdd	327.3 ns/op, 89.3 B/op	294.5 ns/op, 40.8 B/op	-10.0% time, -54.3% alloc
forEachScan	370.8 us/op, 1477.0 B/op	277.6 us/op, 815.4 B/op	-25.1% time, -44.8% alloc
removeAllUpToSmallPrefixAndRefill	19.7 us/op, 6622.3 B/op	18.5 us/op, 4235.4 B/op	-6.2% time, -36.0% alloc

The main effect is lower retained heap from removing one value object per pending entry, plus lower allocation on update/remove paths that no longer need to create replacement pair values. The legacy pair-returning methods remain available for compatibility.

JOL GraphLayout.totalSize() on the value-layout change alone:

Entries / ledgers	Before this PR	This PR	Change
1,000 / 1	64,872 B / 2,007 objects	40,880 B / 1,007 objects	-37.0% size, -1,000 objects
1,000 / 20	73,536 B / 2,083 objects	49,696 B / 1,083 objects	-32.4% size, -1,000 objects
50,000 / 1	3,200,872 B / 100,007 objects	2,000,880 B / 50,007 objects	-37.5% size, -50,000 objects
50,000 / 20	3,209,536 B / 100,083 objects	2,009,696 B / 50,083 objects	-37.4% size, -50,000 objects

Verifying this change

• ./gradlew :pulsar-broker:test --tests org.apache.pulsar.broker.service.PendingAcksMapTest --max-workers=1 -PtestRetryCount=0
• ./gradlew :pulsar-broker:checkstyleMain :pulsar-broker:checkstyleTest --max-workers=1
• ./gradlew :microbench:compileJava --max-workers=1
• ./gradlew :microbench:checkstyleMain --max-workers=1
• ./gradlew :microbench:shadowJar --max-workers=1
• JMH command listed above

[void-ptr974]

After this PR is merged, the next PR will switch the per-ledger entry map in PendingAcksMap to the primitive Long2LongOpenHashMap added in #26028.

That follow-up will keep the packed-value shape from this PR and focus only on removing the boxed inner-map key/value overhead. The BitSet prefix index for same-ledger removeAllUpTo cleanup will be a separate PR after that.

[lhotari]

Related to #26028 (review). I have the fastutil minification working now and this change can be implemented on top of fastutil once the restoration of fastutil has been merged.

[void-ptr974]

Thanks @lhotari. I updated this on top of #26032.

The latest version uses fastutil and keeps the scope to the value-layout change. I also updated the benchmark/JOL numbers in the PR description.