[Bug] DashboardCollectTask.fetchBrokerRuntimeStats discards the retry result, making its retry logic ineffective

### Describe the Bug

`DashboardCollectTask.fetchBrokerRuntimeStats(brokerAddr, retryTime)` retries on failure, but the retry is ineffective: the catch block calls `fetchBrokerRuntimeStats(brokerAddr, retryTime - 1)` **discarding the return value**, then unconditionally rethrows the original exception. Even when a retry succeeds, its result is thrown away and the caller still sees the original failure — the retries only add sleep time.

The existing test actually documents the broken behavior: `DashboardCollectTaskTest.testCollectBroker` stubs the fetch to "throw once, then return a kvTable" (clearly intending to exercise a successful retry), yet asserts that `collectBroker` throws.

### Steps to Reproduce

Stub `fetchBrokerRuntimeStats` to fail once then succeed (as the existing test does); `collectBroker` still fails with `RuntimeException: fetchBrokerRuntimeStats exception`.

### What Did You Expect to See?

A transient failure is absorbed by the retry and the collected stats are returned.

### What Did You See Instead?

The retry result is discarded and the original exception is rethrown unconditionally.

### Additional Context

Fix incoming: `return fetchBrokerRuntimeStats(brokerAddr, retryTime - 1);` in the catch block. When retries are exhausted the method already returns null at `retryTime == 0`, and the caller `collectBroker` already handles it with `if (kvTable == null) continue`.

Start with DashboardCollectTask.fetchBrokerRuntimeStats and DashboardCollectTaskTest.testCollectBroker, which stubs one failure followed by a successful kvTable result. Verify that the retry's returned stats reach collectBroker and update the test expectation so it passes on transient recovery while preserving the retry-exhaustion behavior described in the issue.

Written by the indexing model from the issue text.