[songgaoye]

What problem does this PR solve?

Issue Number: resolve

Problem Summary:

PrometheusMetrics.sanity waits for asynchronous server method metrics to be
exposed after Server::Start().

The test used a fixed number of bthread_usleep() calls. Since
bthread_usleep() may return early when interrupted, the loop can finish in a
few milliseconds instead of waiting for its intended five seconds. This makes
the test flaky when UpdateDerivedVars has not yet exposed the method bvars.

What is changed and the side effects?

Changed:

• Replace the fixed retry count with a five-second deadline based on
  butil::cpuwide_time_us().
• Continue polling until the expected bvar is exposed or the deadline expires.

Side effects:

• Performance effects:

  None. This only affects a unit test.

• Breaking backward compatibility:

  None.

---

Check List:

• Build brpc_prometheus_metrics_unittest
• Run PrometheusMetrics.sanity

[chenBright]

LGTM