[BUG] Thread.sleep(1) + real-clock timestamp ordering assertion

## Description
`Thread.sleep(1)` then `assertThat(second.getLastHeartBeatTime(), greaterThan(firstBeat))` where the timestamp comes from `System.currentTimeMillis()`. On coarse-grained clocks (Windows ~15ms, loaded CI) 1ms does not guarantee a strictly-greater value.

## Location
```
shenyu-admin/src/test/java/org/apache/shenyu/admin/service/impl/InstanceCheckServiceTest.java:105-111
```

## Impact
Intermittent false failure on slow/coarse-clock runners.

## Suggested fix
Inject a controllable clock (or set the field reflectively to a fixed-then-incremented value); assert on the injected values, not wall clock.

## Related existing issue(s)
None

_Identified during the 2026-08-02 audit; full list in [`docs/issue-candidates-2026-08-02.md`](docs/issue-candidates-2026-08-02.md)._

Start at shenyu-admin/src/test/java/org/apache/shenyu/admin/service/impl/InstanceCheckServiceTest.java:105-111 and inspect how the timestamp is obtained from System.currentTimeMillis(). Replace the Thread.sleep(1) timing dependency with the suggested controllable-clock or fixed-then-incremented test setup. Done means the assertion uses deterministic injected values and no longer intermittently fails on coarse-clock or loaded CI runners.

Written by the indexing model from the issue text.