Improve TODOs

[krissetto]

just a note to go improve TODOs at some point.

some thoughts:

• models can leave things behind "in progress" and never finish them;
• we could cleanup done TODOs at some point, either end of request or later on, and provide a way to show the past TODOs;
• maybe introduce minimal dependencies between TODOs, or some form of implicit "get next todo" type of tool so the agent wouldn't have to decide on its own
• etc...

[rumpl]

we could cleanup done TODOs at some point, either end of request or later on, and provide a way to show the past TODOs;

TODOs are cleaned when all are done

[rumpl]

maybe introduce minimal dependencies between TODOs, or some form of implicit "get next todo" type of tool so the agent wouldn't have to decide on its own

I would prefer to create a tasks toolset for this

[dgageot]

I believe both have been implemented