Idea: an optional "Projects and sessions" sidebar layout like all other harnesses

[MarvinFS]

Problem

The left sidebar is a project > worktree tree with agent rows nested under worktree cards. Once you run many agents across several repos, it gets hard to use:

• rows are dense, and the working spinner is small and low contrast
• updated threads are easy to miss while scrolling
• the current row is hard to find again

The Agents view does not fill the gap, because it is built from live panes, not from session history.

Proposal

Add a setting under Settings > Appearance > Sidebar > Sidebar Layout with two values:

• Workspaces (default): today's tree, unchanged.
• Projects and sessions: a flat list.

In the new layout:

• Each project gets a collapsible heading with a +.
• Under it, one line per Claude or Codex session: status marker, title and age, newest first, with "Show more".
• The list covers every session whose working directory belongs to the project, started from Orca or not. It reuses the existing AI Vault scan and its project matching.
• Clicking a live session focuses its pane. Clicking a closed one resumes it in the worktree that owns its folder.
• + starts the default agent in the project folder without creating a worktree. A right-click offers the other detected agents or a plain terminal.
• Worktrees are not a layer: a session in a non-primary worktree shows its branch as a small tag.

Nothing changes for anyone who keeps the default.

Why

This is the model most agent tools have converged on: a project is a folder, and the sessions run against it are listed under it. Orca's worktree tree stays one click away. The header has a toggle back to it, and reveals and renames still open the tree.

Screenshots

Status

I built this as a local patch on v1.4.220 and use it daily on Windows. Type checks, the test suite and the code-quality gate pass.

Known limits:

• local host only; SSH and remote history are not wired up yet
• not yet checked on macOS or Linux
• the first scan of a large history (several GB of transcripts) can take a minute or more, until the parse cache is warm; the list shows a "Reading session history" notice meanwhile

Questions for maintainers

1. Is this a direction you would accept as an optional layout in core?
2. If yes, should SSH and remote support be required before a first PR, or can it land local-only behind the setting?
3. Would you prefer it split into several PRs (setting, data, list, actions)?

Related: #2654, #9509, #15036