fix(server): reconcile blocking result with TaskStore

[jstar0]

Summary

Fixes a race in the blocking Java server response path tracked by a2aproject/A2A#2049. onMessageSend can currently return an internal error when the agent is complete but TaskStore persistence is still becoming visible to the response aggregator.

Root Cause / Context

When the event consumer completes without delivering a Message or full Task event, ResultAggregator performs one TaskStore lookup and immediately raises Could not find a Task/Message if the row is not visible yet. That prevents the later response-handler TaskStore reconciliation from running.

Changes

• Poll TaskStore with a bounded timeout when a blocking call has no in-memory result, while preserving the single-read behavior for non-blocking calls.
• Make the reconciliation timeout configurable through a2a.blocking.reconciliation.timeout.seconds, retaining the existing one-second behavior by default.
• Add debug logging for reconciliation attempts and successful delayed lookups.
• Document the new setting and add regression coverage for delayed visibility, permanent absence, non-blocking behavior, configuration wiring, and the default property value.

Scope / Risk

Only the blocking empty-capture path polls TaskStore, and the default timeout remains one second. EventConsumer lifecycle, TaskStore APIs, cancellation, streaming emission, and the public A2A protocol are unchanged.

Verification

mvn -pl server-common test
mvn -pl integrations/microprofile-config test
mvn -pl server-common,integrations/microprofile-config -am -Dtest=ResultAggregatorTest,DefaultRequestHandlerTest,MicroProfileConfigProviderTest -Dsurefire.failIfNoSpecifiedTests=false test
git diff --check

The module suites pass 416 server-common tests and 10 MicroProfile config tests. The focused cross-module run passes 46 tests.

Fixes a2aproject/A2A#2049

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[kabir]

@jstar0 Thank you! I've opened jstar0#1 with a few fixes to make the timeout configurable and some debug logging.

[jstar0]

Thanks @kabir. I integrated your configurable timeout and debug logging commit, then added coverage for the configuration wiring and default value. The focused 46-test run and the complete server-common (416) and MicroProfile config (10) suites pass.

[kabir]

I don't se the need for the re-introduced constructor. Assuming I'm not mistaken, please get rid of it.
The additional tests look good though 👍