feat: MessageDeliverer.deliverBatch interface + default impl (KOJAK-72)

[endrju19]

Summary

Adds a batch-aware method to MessageDeliverer that transports can override for concurrent I/O. Pure interface introduction with sequential default implementation — zero behavior change today, but unlocks the next two PRs in the optimization chain.

KOJAK-72 — deliverBatch interface

Closes KOJAK-71 — processNext returns Int was folded in here since the same refactor naturally exposes the count.

Changes

MessageDeliverer.deliverBatch(entries)

Default sequential implementation — loops over single-entry deliver(), preserves input order and per-entry result classification. Backward compatible: existing implementations don't need to change.

CompositeMessageDeliverer.deliverBatch

Groups entries by deliveryType, delegates each sub-batch to the matching deliverer's deliverBatch. Fails permanently for unknown types (consistent with single-entry deliver). Results re-assembled in original input order.

OutboxEntryProcessor.processBatch(entries)

New method — calls deliverer.deliverBatch once, applies retry policy per result. Existing process(entry) preserved for single-entry callers (deduplicated via shared applyResult helper).

OutboxProcessor.processNext

• Switches from per-entry forEach { entryProcessor.process(...) } to a single processBatch(claimed) call
• Returns Int (count of processed entries) — subsumes KOJAK-71

Semantic note: per-entry duration in batch path

OutboxProcessingEvent.duration semantics change in the batch path: each entry receives the wall-clock duration of the whole batch, not its individual delivery. This is intentional — transports overriding deliverBatch (Kafka fire-flush-await, parallel HTTP sendAsync) overlap per-entry I/O internally, making per-entry timing meaningless.

MicrometerOutboxListener (KOJAK-44) is unaffected — it uses onBatchProcessed for batch timing and only counts per-outcome events.

Documented in OutboxProcessor KDoc.

Verification

• All unit tests pass (./gradlew test -x :okapi-integration-tests:test)
• All integration tests pass (./gradlew :okapi-integration-tests:test — Postgres/MySQL/Kafka/HTTP via Testcontainers)
• ktlint clean
• Smoke benchmark — Kafka throughput, batchSize=50:
  • Baseline (KOJAK-68): 8.665 ms/op (~115 msg/s)
  • This PR: 8.321 ms/op (~120 msg/s)
  • Difference within measurement noise → zero regression confirmed

What's next

This PR introduces the interface. The actual throughput improvements come when transports override deliverBatch:

• KOJAK-73: KafkaMessageDeliverer.deliverBatch — fire-flush-await pattern, expected 10-100× Kafka throughput improvement
• KOJAK-74: HttpMessageDeliverer.deliverBatch — parallel httpClient.sendAsync, expected 10-50× HTTP throughput at realistic webhook latency

Test plan

• New unit tests: MessageDelivererTest (default deliverBatch behavior), CompositeMessageDelivererTest (grouping + permanent failure for unknown types + override usage)
• Extended OutboxEntryProcessorTest with processBatch cases (mixed results, empty input)
• Extended OutboxProcessorTest with return-value assertions