Speed up TestOnReconnectListenerSupport by porting to SolrCloudTestCase

[janhoy]

The test spun up 4 Jetty nodes (1 control + 3 cloud, sliceCount=2/fixShardCount(3)) via the legacy AbstractFullDistribZkTestBase, plus a control collection — yet only ever creates and inspects a 1-shard/1-replica collection.

Rewritten on SolrCloudTestCase with a single node, using the cloud-managed configset (ManagedIndexSchemaFactory, which is what registers ZkIndexSchemaReader as an OnReconnect listener — the behavior under test). All assertions are preserved:

• ZkIndexSchemaReader registered as an OnReconnect listener for the core
• after collection RELOAD, the old core's listener is unregistered and the new core's is registered
• after collection DELETE, the listener is unregistered (now with an explicit wait instead of a one-shot check)

Measured locally: suite time drops from ~14s to ~3s (test method 13.6s → 0.7s). Also one fewer test on the legacy AbstractFullDistribZkTestBase.

Test-only change, no changelog entry.

Why and how

Crave is shutting down. We need to speed up our tests. I asked claude to pick 50 solr-core tests at random and look for speed improvements. It picked this first and improved it from 13 to 1 seconds.

[dsmiley]

BTW we needn't tell an LLM to find tests to improve "at random"; that seems inefficient. Our test suite prints the slowest tests at the end... and furthermore I strongly suspect Develocity can tell us the slowest ones. We may want to approach this by identifying a theme and tackling that theme to have good focus/context. Perhaps retiring AbstractFullDistribZkTestBase would be a good theme. But I'd recommend beginning with a conversation and not action.

[janhoy]

This is a spike, fixing one random independent test. Not a start of something bigger or a strategy suggestion. It simply proves that just by looking at a few random tests there was lots to gain.

For strategies, one could look into a certain base class, identify polling and sleep patterns etc.

And ad proven in the other PR I put up, some test scenarios, not all, are good candidates for mocks.