Stability improvements for sql/zk usage

[ssalinas]

Few things here after we experienced some hiccups with persisting:

• Sort the task persisting by child node count, to take care of more expensive/troublesome parent nodes first
• Enable spinning up a separate instance that only serves api calls (i.e. a deployment pattern to isolate the scheduler)
• Optimize the active tasks for request id check
• make a cheaper HEAD request method to check if a request exists without having to fetch full request + active task data for the request parent object

[pschoenfelder]

🚢 looks good to me

[mikegajda]

🚢 LGTM