Potential deadlock in index writer threads under heavy load

Originally reported on EAP: https://issues.redhat.com/browse/EAPSUP-961

In some cases, it seems Infinispan will try to follow-up an indexing operations by another.

See https://github.com/infinispan/infinispan/blob/7bd0ed0b68eaa48b8b6e15496316da54914cc895/query/src/main/java/org/infinispan/query/backend/QueryInterceptor.java#L433

The request to update indexes is added in thenCompose(), which means it's added as soon as the first operation is marked as complete, which actually happens in the index writer thread... which is precisely the thread responsible for consuming that queue.

That can lead to a deadlock: should the queue be full, the act of adding an element to that queue will block, which means the index writer thread, responsible for consuming that queue, will block, which means the queue will never get processed ever again.

The bug seems to have been introduced in ISPN-11731: https://github.com/infinispan/infinispan/commit/c68c255f3b81dfcb71a25051b38be94700ad5ef5