Fix issue with stalled requests if shutdown

[kbirk]

I believe this to be the cause of the stalled requests that @mwdchang experienced.

If the hmi-server was shutdown between writing the initial TaskStatus.QUEUED state of the task to redis but before it actually sent the request to rabbitmq, then any subsequent request would simply wait for a non-existent task to finish.

This is consistent with the logs as no request was sent, but it did timeout waiting for the response.

This switches the logic to write out the TaskStatus.QUEUED state after it is dispatched. If the server is shutdown after it dispatches to redis but before that state is set, then the subsequent requests will simply re-send it.

Another small issue is that failed task id's were never re-written in the task SHA -> task id lookup. So after a task failed subsequent attempts, if successful, were not cached.