Add Javascript Orchestration for Chatbot/Assistant

Problem/Motivation

Currently you ask the chatbot to do something and you have to wait anything from a short time to very long time to get a response. Its very unclear for both a developer or a informal user what happens during this time.

In the end you can have either a hallucinated or hard to understand responses from the assistant saying that its currently working on something or that it will work on something or even that it has done something that is not actually in the response from the agents, causing frustration for the end user.

Most popular AI chatbots like ChatGPT or Claude have visual verification when its using tools like searching the web or calculating something, so you can understand that it did something. In the end the Agents are just tools for the Assistant and we should have a similar UX.

The Assistants API is already half prepared for this and you also get the benefit of having javascript orchestrate the agents and subagents, so you have lower chance of running into timeout problems.

Proposed resolution

• Add so that the action plugins for the Assistant can be split up into steps.
• Add so that the Assistants can trigger each of the steps that is happening and send feedback with title, description and meta data if its finished.
• Add so that the cleanup function still works, so we don't get stuck with a half-setup configurations if its an agent or that it does RAG searches in the background that are not needed.
• Add so that the Deepchat chatbot can trigger each step autonomously.

User interface changes

• Add design/graphics for how a tool working screen looks like.

API changes

• Action plugins needs an method to declare an end or continue state after they are done and what the metadata of the next step is.