Is your feature request related to a problem?
Followup to #4857, #4886 -- if a query has a very long timeout and is unexpectedly taking longer than expected, we should be able to cancel queries via the _tasks/_cancel API.

What solution would you like?
For the most part cancellation should already be supported fundamentally by #4857, we just need to integrate with the actual task endpoint and let that interrupt the query thread. We may also need to support canceling whatever background IO is running during that process (but I think this will self-interrupt anyway, since the socket will get closed & OpenSearch detects that).

What alternatives have you considered?
N/A

Do you have any additional context?
N/A