Removes all joinAndUnwrap calls from ClaimTask, SplitTask and MergeTask.
Each task's internal logic is now a proper async chain — no coordinator
threads are blocked waiting on token store operations.

Key changes:
- ClaimTask: two sequential UoW calls chained with thenCompose; fetchToken
  failure caught via exceptionally returning false (preserving original
  try/catch behavior)
- SplitTask: splitAndRelease(Segment) changed from boolean to
  CompletableFuture<Boolean>; callers updated from thenApply to thenCompose
  to avoid unwaited nested futures; try/finally replaced with whenComplete
  to guarantee deadline cleanup on both success and failure
- MergeTask: two sequential segment fetches chained with thenCompose;
  thenCombine(...).thenCompose(f -> f) keeps parallel token fetching while
  flattening the nested future; mergeSegments now returns
  CompletableFuture<Boolean> with whenComplete for deadline cleanup

Adds ClaimTaskTest covering all four scenarios: segment already active,
segment not in available list, successful claim, and failed claim (returns
false as normal completion, not exceptional).

Extends MergeTaskTest with a missing branch: merge initiated from the
higher-ID segment, verifying the lower-ID segment token is always placed
first in MergedTrackingToken regardless of which segment triggered the merge.

Refs #3773

Replace all blocking joinAndUnwrap call sites in WorkPackage (4 sites)
and Coordinator (8 sites) with proper async chains using thenCompose,
thenAccept, and CompletableFuture composition.

Key changes:
- WorkPackage: storeToken, extendClaimIfThresholdIsMet, processEvents,
  and scheduleWorker now return CompletableFuture<Void> instead of blocking
- Coordinator: start() returns CompletableFuture<Void>; initializeTokenStore
  becomes async with retry logic; claimNewSegments, createWorkPackage,
  ensureOpenStream, and extendClaim handling all converted to async chains
- CoordinatorTest: comprehensive tests for all new async behaviour including
  stale-task generation guard, extend-claim failure handling, claim-token
  error propagation, onSegmentReleased lifecycle, start() state transitions,
  and initializeTokenStoreWithRetry retry/exhaustion paths

…semantics

Add processor lifecycle documentation to reflect that start() is fully
async and only completes after the token store identifier is resolved and
the coordinator has started.

- index.adoc: add general lifecycle section covering start(), shutdown(),
  isRunning(), and isError() for all EventProcessor implementations
- streaming.adoc: add PSEP-specific lifecycle section before Configuring,
  documenting the two-step start() sequence and the IllegalStateException
  thrown by getTokenStoreIdentifier() when called before start() completes
- streaming.adoc: update "Retrieving the token store identifier" section
  to reflect eager resolution during start()

…ugfix

[#3773]: phase 0 Replace thenRun + joinAndUnwrap with thenCompose in Coordinator#abortWorkPackage

…ronously

Move mergedWith and MergedTrackingToken.merged inside executeWithResult
so synchronous exceptions are captured by UnitOfWork.safe() and
whenComplete always removes releasesDeadlines entries even on failure.

Fix thenThrow → failedFuture in MergeTaskTest and ClaimTaskTest for
CompletableFuture-returning mocks to correctly simulate async failures.

Add WhenMergeSegmentsFails test verifying releasesDeadlines are cleared
when mergeSegments throws due to incompatible segments.

Refs #3773

[#3773]: phase 1 Make ClaimTask, SplitTask and MergeTask fully async

…ge-coordinator

[#3773]: phase 2 remove joinAndUnwrap from workPackage and coordinator

…e 3)

Resolve the token store identifier eagerly during start() instead of lazily
blocking on the first call to getTokenStoreIdentifier().

Changes:
- start() now chains tokenStore.retrieveStorageIdentifier() before
  coordinator.start(), forming a fully async pipeline with no blocking
- calculateIdentifier() helper removed along with its joinAndUnwrap call
- AtomicReference<String> replaced with a volatile String field; a single
  write in start() and read-only access afterwards makes AtomicReference
  unnecessary
- getTokenStoreIdentifier() is now a direct field lookup; throws
  IllegalStateException when called before start() has completed, making
  the lifecycle contract explicit instead of silently blocking
- Updated existing getTokenStoreIdentifier test to call startEventProcessor()
  first; added test for the pre-start IllegalStateException

- Resolve identifier lazily in getTokenStoreIdentifier() instead of throwing
  IllegalStateException, using a dedicated resolveTokenStoreIdentifier() method
  with FutureUtils.joinAndUnwrap
- Update StreamingEventProcessor interface Javadoc to accurately describe
  both lazy and eager resolution strategies
- Update PSEP override Javadoc to reflect new lazy fallback behaviour
- Add tests for lazy resolution (happy path and failure propagation)

… lifecycle snippets

Update the three code snippets flagged in review to inject AxonConfiguration
and retrieve the processor via getComponent(Class, name) before using it,
matching the real-world usage pattern shown in the framework examples.

[#3773] Document PooledStreamingEventProcessor lifecycle and start() …

…tifier

[#3773]: phase 3 Remove joinAndUnwrap from PooledStreamingEventProcessor

…rocessor

# Conflicts resolved in:
#	messaging/src/main/java/org/axonframework/messaging/eventhandling/processing/streaming/pooled/PooledStreamingEventProcessor.java
#	messaging/src/main/java/org/axonframework/messaging/eventhandling/processing/streaming/pooled/WorkPackage.java
#	messaging/src/test/java/org/axonframework/messaging/eventhandling/processing/streaming/pooled/WorkPackageTest.java