Solidify contract that all accepted points are processed

In the current implementation when the server is shutdown in-flight requests are dropped.

The correct design should be that all tasks run to completion but to reload a task its source is cut off, and is then allowed to run to completion. Once all tasks run to completion we need to add an explicit abort method that can kill a running task without allowing it to run to completion.

This is going to need some careful design and API specification.

No files, tests, or entry points are named. Begin by mapping shutdown, reload, and in-flight task handling, then define the lifecycle and API behavior for completion versus explicit abort; done means the design is specified and accepted.

Written by the indexing model from the issue text.