Make Agent tool calls compact and easy to recognize

[marksftw]

Summary

Make Agent Mode tool calls take up less space and make common tool types easier to recognize, without moving or hiding any calls.

Scope

• Keep every tool call in its current timeline position.
• Make the default tool row more compact while preserving its name, status, error state, and expandable details.
• Use consistent icons for tool types that can be identified reliably from existing UI data, such as shell commands, file reads, file edits or writes, web activity, and MCP tools.
• Use a clear generic icon for unknown or unsupported tool types. Do not depend on brittle parsing of commands or output text.
• Keep text labels and accessible names so meaning is not conveyed by icons alone.
• Preserve current approval behavior and failed-tool visibility.

Done when

• Tool-heavy timelines are easier to scan and use less vertical space.
• Users can still inspect the same inputs, outputs, statuses, and errors as before.
• The layout works in compact, desktop, light, and dark presentations.

Out of scope

• Reordering or hiding tool calls
• Grouping multiple calls together
• Changing the underlying timeline or backend data

[AnthonyRonning]

Implemented and merged in #709. The scope of this issue is honored as written: Agent tool rows are now more compact and easier to recognize in bulk, while every call remains in place with its status, failure treatment, approvals, and expandable details. This is a strict improvement to the existing presentation.

The resulting tool-heavy view is intentionally still the same basic one-row-per-call design, and it remains too text-heavy and full-width to be the long-term activity UI. #712 now documents the separate desired redesign: grouped activity phases, progressive disclosure, quiet completed summaries, prominent running/approval/failure states, concise semantic targets, full detail recovery, and accessible light/dark/narrow behavior.