Fix Windows orchestrator test race condition

[circleci-app]

Summary

This PR fixes a race condition in the orchestrator tests that was causing failures on Windows, specifically in the test case "error: task agent encountered fatal error".

Problem

The test was failing on Windows with the following error:

=== FAIL: task TestOrchestrator/error:_task_agent_encountered_fatal_error (48.50s)
    orchestrator_test.go:327: assertion failed: 
        --- ←
        +++ tt.wantTaskEvents
          []fakerunnerapi.TaskEvent(
        -       nil,
        +       {
        +               {
        +                       Allocation:     "testalloc",
        +                       TimestampMilli: 1779310915952,
        +                       Message:        []uint8("error while executing task agent: task agent command exited with"...),
        +               },
        +       },
          )

The orchestrator was unable to send the failure event to the fake runner API, with connection refused errors occurring across 16 retry attempts over 48 seconds.

Root Cause

On Windows, there's a race condition between:

1. The orchestrator's asynchronous error handling (which sends failure events to the runner API)
2. The test's synchronous assertions (which immediately check for those events)

Windows uses Job Objects for process cleanup, which is synchronous and creates tight timing windows. The test was asserting before the orchestrator had completed sending the failure event.

Solution

This PR adds polling logic to wait for the expected number of task events and unclaims (with a 30-second timeout) before performing the assertions. This approach:

• Eliminates the race condition by waiting for async operations to complete
• Still catches genuine failures (via the timeout)
• Doesn't slow down tests that complete quickly (polling succeeds immediately when data is available)

Testing

• Test compiles successfully
• Solution mirrors the fix from commit 7b927ad which addressed the same issue

Original failing job
Agent run