.NET: Forward Magentic participant replies to manager

[ghominejad]

Fixes #6172 , #6173

Motivation and Context

The .NET MagenticOrchestrator does not feed the participant agents'
responses back into the manager's chat history. The manager builds its
progress-ledger prompt from MagenticTaskContext.ChatHistory, so it
never sees any participant replies and keeps re-picking the same speaker
until MaxRounds terminates the run.

The Python port handles this in _MagenticOrchestratorExecutor._handle_response,
which extends magentic_context.chat_history with the participant's
messages before re-entering the inner loop. This PR restores parity.

with a two-participant workflow where both participants reply with a short status token, the Python sample terminates in 3 rounds with the expected verdict, while .NET loops until MaxRounds without ever incorporating either reply.

Description

In MagenticOrchestrator.TakeTurnAsync, the messages parameter on
subsequent turns carries the participant agent's reply (forwarded by
AIAgentHostExecutor and accumulated by ChatProtocolExecutor). The
first-turn branch consumes messages when constructing the initial
MagenticTaskContext, but the subsequent-turn branch ignored it and went
straight to RunCoordinationRoundAsync, so the reply was dropped.

The fix appends the incoming messages to _taskContext.ChatHistory
before invoking the coordination round, mirroring the Python orchestrator's
chat_history.extend(messages) step. No public API change, no behavior
change on the first turn or on terminated contexts.

Tests

• Added RunCoordinationRound_Forwards_Participant_Reply_To_ManagerAsync in
  MagenticOrchestrationTests — runs a 2-round scenario and asserts the
  worker's reply appears in the manager's round-2 progress-ledger input.
  Verified to fail without the fix.
• Added RecordingReplayAgent test helper (sibling of TestReplayAgent)
  that captures input messages per call, for any future test that needs to
  assert what the manager was actually handed.
• All MagenticOrchestrationTests pass

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible
• Is this a breaking change? If yes, add "[BREAKING]" prefix to the title of the PR.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Fixes a bug in MagenticOrchestrator.TakeTurnAsync where participant replies on subsequent turns were dropped instead of being appended to the manager's chat history, causing the orchestrator to repeatedly dispatch the same speaker until MaxRounds.

Changes:

• Append incoming messages to _taskContext.ChatHistory before running coordination rounds on subsequent turns.
• Add a RecordingReplayAgent test helper that captures input messages on each call.
• Add a regression test verifying round-2 progress ledger sees the worker's reply.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
dotnet/src/Microsoft.Agents.AI.Workflows/Specialized/Magentic/MagenticOrchestrator.cs	Forwards participant reply messages into the manager's chat history on subsequent turns.
dotnet/tests/Microsoft.Agents.AI.Workflows.UnitTests/RecordingReplayAgent.cs	New test agent that records inputs while replaying scripted responses.
dotnet/tests/Microsoft.Agents.AI.Workflows.UnitTests/MagenticOrchestrationTests.cs	Adds regression test asserting participant replies reach the manager.

[ghominejad]

@microsoft-github-policy-service agree