Group Agent reasoning and tool calls into collapsible steps

[marksftw]

Summary

Group related Agent reasoning and tool calls into collapsible steps so longer timelines are easier to scan.

Scope

• Build on the descriptive labels from Give completed Agent thoughts descriptive labels #646, their task-history persistence from Persist descriptive Agent thought labels with task history #647, and the tool-row improvements from Make Agent tool calls compact and easy to recognize #648.
• Keep the underlying Agent timeline flat. Derive grouping only for presentation unless a small supporting data change is required for a reliable boundary.
• Start a step for each visible reasoning phase and use its descriptive label as the step name. Following tool calls belong to that step until the next reasoning phase or final response.
• Put tools without visible reasoning in a fallback step named Working.
• Keep the active step open by default so running tools remain visible. Older steps may default collapsed, while preserving the user's manual open or closed choice.
• Keep reasoning details and individual tool details independently accessible inside an open step.
• Keep approvals, system notices, errors, and final assistant responses outside steps, visible, and in their original order.
• Define the simplest reliable behavior for intermediate assistant text and for standalone cards that interrupt a step. Do not reorder timeline activity to make grouping easier.

Done when

• A user can scan descriptive step names and expand any step to inspect its reasoning and tool activity.
• Active work and approvals remain visible and actionable.
• Grouping does not change stored timeline content or the order of events.

Out of scope

• Redesigning approvals
• Rewriting or removing raw reasoning
• Changing normal Chat reasoning UI