fix(adka2a): exclude artifacts from input-required A2A responses

[anishesg]

Link to Issue

Fixes #913

Problem

When adk-go sends A2A responses for input-required tasks, it includes artifacts containing messages alongside the status update. This format diverges from adk-python's implementation, which sends only a rich history structure without artifacts. The artifacts section triggers incorrect behavior in adk-python clients connecting via RemoteA2aAgent, causing them to generate a spurious mock_function_call_for_required_user_input.

Solution

Modified the A2A response serialization in server/adka2a/v2/input_required.go and server/adka2a/v2/executor.go to ensure that for input-required tasks, all content (function calls, responses, and text messages) is included in the status message instead of being sent as separate artifacts. This aligns adk-go's response format with adk-python's expectations.

The fix includes two key changes:

• In input_required.go, the inputRequiredProcessor now accumulates all parts from events after detecting a long-running function call, placing them in the status message rather than leaving them to be sent as artifacts.
• In executor.go, the writeFinalTaskStatus method now skips sending the partial artifact reset when the final state is TaskStateInputRequired.

Testing Plan

Unit Tests

• Updated and ran unit tests in server/adka2a/v2/processor_test.go
• All existing tests pass with the updated behavior
• Test case "long running tool call before receiving all parts" now correctly expects text messages in the status message instead of artifacts

=== RUN   TestEventProcessor_Process
--- PASS: TestEventProcessor_Process (0.01s)
    --- PASS: TestEventProcessor_Process/long_running_tool_call_before_receiving_all_parts (0.00s)
PASS
ok  	google.golang.org/adk/server/adka2a/v2	0.031s

[anishesg]

checked the CI failure - all tests actually pass, including the modified test case. the bare "FAIL" at the end without a package name or error message looks like a CI infrastructure glitch. local runs with the same flags pass cleanly. can we re-run the workflow?