Provide a buffer initialization stage

[mxgrey]

Sometimes it helps if buffers can be initialized with certain values before a workflow begins running. A user can manually design the workflow in a way that ensures the buffers are initialized before any other node gets run, but this may take some careful thought on the user's part.

The initialization stage would be the complement of the cleanup stage. Where the cleanup stage runs after a workflow has finished running, the initialization stage would run before. Similar to cleanups, initializations are themselves workflows whose initial inputs are one or more buffers and whose final output is a trigger.

The only significant challenge is that in many cases initialization workflows will want to see the input value of the parent workflow. It is not clear how to generalize that visibility in a sound way. We will probably need to rely on InRef which was introduced in a more recent version of bevy.

[koonpeng]

I think we could just have a node/service that pushes the initial data into the buffer. Not sure if it is an issue for init data that does not require the scope input since nothing will "trigger" it, but maybe services solves that.

[mxgrey]

You're absolutely correct that a user can already manually set up the workflow so that the buffers get filled in before the main workflow begins.

However, as I start getting some practice with writing real workflow examples I'm seeing a recurring pattern where I want to set up a bunch of buffers with initial data and ensure the values are set before the real heavy lifting of the workflow begins. To do that I need to trigger various branches at startup to fill in the buffers and then have a listener that watches for all the buffers to be set before the real workflow begins.

I think the pattern is common enough (and the manual way of doing it is clunky enough) that it will be worth designating a startup stage like this. We already have the cleanup stage as a model for this, and we can just duplicate that procedure but put it at the workflow startup instead of teardown.

This will also make it easier for us to support things like an initial_values: [ ... ] field for buffer operations in diagrams.