#7760: fix(agents): resolve message ordering conflict during tool execution

## Problem When a user sends a new message while the agent is actively executing tool calls, a "Message ordering conflict" error is triggered. The bot appears to work but the user's message is lost, requiring them to resend after the agent finishes. Error message: ``` Message ordering conflict - please try again ``` ## Root Cause `clearActiveEmbeddedRun()` signals "run ended" before `flushPendingToolResults()` writes tool results to the session file. Since: ``` clearActiveEmbeddedRun() → waiters proceed → read incomplete session → role ordering error ``` The incoming message sees an incomplete transcript (missing tool results), causing consecutive user messages which violates role alternation rules. ## Fix Move `clearActiveEmbeddedRun()` to after `flushPendingToolResults()` in the outer finally block. This ensures tool results are persisted before any waiting messages proceed. ## Impact **Before:** User messages during tool execution trigger ordering conflicts. Message is lost. **After:** Tool results are flushed first, then waiters proceed. No conflict. Fixes #7694 <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR adjusts the embedded Pi runner teardown sequence in `src/agents/pi-embedded-runner/run/attempt.ts` to prevent “message ordering conflict” errors when a new user message arrives while tool calls are still being flushed. The main change is moving `clearActiveEmbeddedRun()` out of the inner cleanup and into the outer `finally`, after `sessionManager.flushPendingToolResults()` so that any waiter that proceeds will see a complete transcript (including synthetic tool results) before starting a follow-up run. The change fits into the existing concurrency model where (1) the session write lock protects session file writes and (2) the active-run registry (`ACTIVE_EMBEDDED_RUNS`) gates whether messages can be queued / whether callers wait for a run to finish. <h3>Confidence Score: 4/5</h3> - This PR is likely safe to merge and addresses a real race, with low residual risk. - The change is localized and aligns the active-run signaling with persistence of tool results, which should prevent incomplete transcripts from being read by concurrent follow-up messages. Main risk is around ensuring all execution paths that register an active run also clear it; the new `queueHandle` guard makes that relationship a bit more implicit and worth double-checking. - src/agents/pi-embedded-runner/run/attempt.ts <!-- greptile_other_comments_section --> _{(4/5) You can add custom instructions or style guidelines for the agent [here](https://app.greptile.com/review/github)!} **Context used:** - Context from `dashboard` - CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=fd949e91-5c3a-4ab5-90a1-cbe184fd6ce8)) - Context from `dashboard` - AGENTS.md ([source](https://app.greptile.com/review/custom-context?memory=0d0c8278-ef8e-4d6c-ab21-f5527e322f13)) <!-- /greptile_comment -->