Wish: Programmatic workflow management

[pringshia]

It may be nice to have a programmatic approach to designing workflows for a task. Here's an example:

# when a worker accepts a task, a stateful workflow
# method gets called, which yields task data

async def workflow(state):

  state.set("role", lambda: random("speaker", "listener"))

  # throws if worker fails onboarding, no more tasks sent:
  assert await onboarding_task()
  for x in range(3):
    for n in range(10):
      await task(state.role)
    # throws if worker fails gold task, no more tasks sent:
    assert await gold_task()

Some implementation notes

The workflow method is designed to be resumable. This means that it needs to keep track of state. It also means that "side-effects" are cached. For example, the first time a user accepts a HIT, the await suspends for onboarding_task. The next time a user accepts a HIT, the await is instantaneous for onboarding_task because the workflow fn is state-aware and caches the previous result.

• Open question: do we allow users to modify the file in between runs? That could mess things up big time, and cause inconsistencies