concurrency=1000,reserveN=10,limit=400000 is flaky

## Flaky Test

### Which jobs are failing
```
QPS: 42597.00, RU: 425970.00, new request need wait 0s
--- FAIL: TestQPS (4.11s)
limiter_test.go:232:
Error Trace: /home/prow/go/src/github.com/tikv/pd/client/resource_group/controller/limiter_test.go:232
Error: "25970" is not less than or equal to "1000"
Test: TestQPS
--- FAIL: TestQPS/concurrency=1000,reserveN=10,limit=400000 (4.11s)
FAIL github.com/tikv/pd/client/resource_group/controller 11.791s

QPS: 45770.00, RU: 457700.00, new request need wait 0s
--- FAIL: TestQPS (4.11s)
limiter_test.go:232:
Error Trace: /home/prow/go/src/github.com/tikv/pd/client/resource_group/controller/limiter_test.go:232
Error: "57700" is not less than or equal to "1000"
Test: TestQPS
--- FAIL: TestQPS/concurrency=1000,reserveN=10,limit=400000 (4.11s)
FAIL github.com/tikv/pd/client/resource_group/controller 11.787s
```

### CI link
https://prow.tidb.net/view/gs/prow-tidb-logs/pr-logs/pull/tikv_pd/10639/pull-unit-test-next-gen-3/2051808173110595584
https://prow.tidb.net/view/gs/prow-tidb-logs/pr-logs/pull/tikv_pd/10632/pull-unit-test-next-gen-3/2051858624984125440

### Reason for failure (if possible)
Retained as flaky: the same subtest failed on two different PRs (10632 and 10639) with the same limiter_test.go:232 assertion and near-identical raw log shape in both ci_links and log_ref logs; PR 10639 does not touch client code, PR 10632 does not touch client/resource_group/controller, and 10 local reruns passed, so this is more consistent with a timing-sensitive test than a commit-specific regression.

### Anything else
Observed in `pull-unit-test-next-gen-3` on PR #10632 and PR #10639.