[BUG] _msearch API doesn't properly handle task cancellation

[msfroh]

Describe the bug

The _msearch API will execute some number of requests concurrently (up to max_concurrent_searches), but will queue up any requests beyond that and execute them in a callback once initial requests complete.

If the task times out (or is otherwise canceled), we still try to execute the subsequent requests anyway. This throws an uncaught exception with the following stack trace:

TaskCancelledException[The parent task was cancelled, shouldn't start any child tasks, channel closed]
at org.opensearch.tasks.TaskManager$CancellableTaskHolder.registerChildNode(TaskManager.java:680)
at org.opensearch.tasks.TaskManager.registerChildNode(TaskManager.java:350)
at org.opensearch.action.support.TransportAction.registerChildNode(TransportAction.java:78)
at org.opensearch.action.support.TransportAction.execute(TransportAction.java:97)
at org.opensearch.client.node.NodeClient.executeLocally(NodeClient.java:112)
at org.opensearch.client.node.NodeClient.doExecute(NodeClient.java:99)
at org.opensearch.client.support.AbstractClient.execute(AbstractClient.java:480)
at org.opensearch.client.support.AbstractClient.search(AbstractClient.java:611)
at org.opensearch.action.search.TransportMultiSearchAction.executeSearch(TransportMultiSearchAction.java:180)
at org.opensearch.action.search.TransportMultiSearchAction$1.lambda$handleResponse$0(TransportMultiSearchAction.java:200)
at org.opensearch.common.util.concurrent.ThreadContext$ContextPreservingRunnable.run(ThreadContext.java:964)

This results in a "zombie" indices:data/read/msearch task.

Related component

Search:Resiliency

To Reproduce

N/A

Expected behavior

If the parent msearch task is canceled, we should not try to run any more search requests.

Additional Details

N/A