Concurrent load is too vulnerable to slow extracts

[bhtucker]

If extracts stop for an hour, a concurrent load will exit with a ETLRuntimeError, regardless of whether the relations it is waiting on extracts for are required.

[tvogels01]

The first approach here would be to have "failure" events sent into the load queue, leaving it up to the load workers to discover whether the relation is required or not.

The concurrent load bears some refactoring with the monitor package. An idea here is that this pseudo-event of a failed extract (= failed due to timeout) is added to the queue where the monitor querier is the producer and the concurrent load is the consumer.