Faulty workflow.yaml can be saved from AI chat

@elias-ba recently pointed out that the workflow chat agent can send an invalid workflow.yaml to Lightning and we'll render it without question. This has lead to the generation of some broken workflows. (Namely, one with multiple "first nodes" all stemming directly from the trigger.)

Saving a workflow like that would lead to all sorts of problems. We should validate the workflow.yaml that's being sent back from the AI assistant before rendering it. And if it's invalid, ideally, the workflow agent would get to see the error and try again.

On the workflow importer, we have a function that validates the proposed worklfow.yaml before attempting to render it on the canvas. Maybe we can use some of this?

Version number v2.13.1

I have reproduced this locally on main:

• Yes
• No

To Reproduce Steps to reproduce the behavior:

1. Go to the workflow chat
2. Ask it to generate a job with lots of different "initial nodes" stemming directly from the cron trigger.

--

P.S., @elias-ba , are you certain we don't have an issue for this? I couldn't find one either, but I swear that @josephjclark , @hanna-paasivirta, or @stuartc was talking about an Apollo-side call to a validation service and I'd be surprised if there wasn't an issue somewhere for checking the generated workflow.yaml before sending it back to lightning—possibly even with an externally available "workflow validator API" or something like that.

Start with the workflow chat entry point and the workflow importer validation function mentioned in the issue, then trace where the generated workflow.yaml is rendered on the canvas. Done means invalid workflows are rejected before rendering and the validation error is made available to the workflow agent for another attempt.