Test/2026 05 12 lark bot reply chain regressions

[YueZh127]

Summary

Part of #633.

This change adds focused regression coverage for the lark-bot reply chain on feature/lark-bot, covering the current follow-up scope from:

• [test] 补齐 AgentRunGAgent 与 conversation->run handoff 的 actor 语义回归测试 #635 AgentRunGAgent terminal / stale / duplicate regressions
• [test] 补齐 TurnStreamingReplySink 的并发收尾回归测试 #636 TurnStreamingReplySink targeted race regressions
• [test][guard] 补齐 reply generator / tool closeout 回归测试并增加最小结构护栏 #637 dispatcher seam guard + generator warning / closeout basics

What changed

#635 AgentRunGAgent regression coverage

Added basic combination regressions around terminal and cleanup behavior:

• ignore cleanup for non-terminal runs
• ignore cleanup when AgentRunCleanupRequested.RunId does not match the terminal run
• do not dispatch duplicate LlmReplyReadyEvent after a terminal failed fallback reply has already been produced and delivered

These assertions are aligned with the current contract:
terminal failure after successful fallback delivery remains
ReplyProduced + ReplyDispatched + ProducedTerminalState=Failed,
rather than Status=Failed.

#636 TurnStreamingReplySink targeted race regressions

Added two sink-level race regressions:

• Dispose() while FinalizeAsync() is waiting for drain unblocks correctly and does not dispatch the stashed final flush
• if a deferred timer-driven flush dispatch fails, a later delta can still recover and publish the latest accumulated text once

This keeps the sink coverage focused on real race-prone paths instead of broad coverage expansion.

#637 dispatcher seam + generator basics

Added:

• a minimal architecture/source-text guard that keeps ConversationGAgent on the IChannelLlmReplyRunDispatcher seam and prevents direct references to concrete run/inbox runtime types
• generator coverage for:
  • warning behavior when SkillRegistry is present without IRemoteSkillFetcher
  • placeholder + tool-follow-up path, with direct proof that a second request containing a tool message is issued

Also tightened naming/intent so the tests describe the current verified contract, not a stronger target-state contract.

Audit doc update

Updated the reply-chain audit document to reflect the current test reality:

• ChatRuntime terminal chunk behavior is still a target-state question, not an accepted current contract
• the dispatcher seam guard is currently a minimal source-text guard, not a Roslyn/compile-level rule
• the new sink regressions are recorded as the right next-step coverage for this branch

Verification

• test/Aevatar.GAgents.ChannelRuntime.Tests
  filtered to AgentRunGAgentTests | ConversationReplyGeneratorTests | TurnStreamingReplySinkTests
• test/Aevatar.Architecture.Tests
  filtered to the dispatcher seam guard test
• bash tools/ci/test_stability_guards.sh

Results:

• channel runtime targeted tests: passed
• architecture targeted test: passed
• stability guard: passed

Notes

This change intentionally does not modify ChatRuntime production behavior.
It only adds/adjusts tests and audit documentation against the currently accepted behavior on this branch.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.50%. Comparing base (806b181) to head (f603372).
⚠️ Report is 67 commits behind head on feature/lark-bot.

@@                 Coverage Diff                 @@
##           feature/lark-bot     #651     +/-   ##
===================================================
  Coverage             72.50%   72.50%             
===================================================
  Files                  1286     1296     +10     
  Lines                 95060    96507   +1447     
  Branches              12428    12632    +204     
===================================================
+ Hits                  68920    69972   +1052     
- Misses                21179    21485    +306     
- Partials               4961     5050     +89     

Flag	Coverage Δ
ci	72.50% <ø> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 48 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.