fix(sidebar): sleeping sessions obey the group filter

[mortenaslo]

Note

@AThraen: I have not tested this manually in the running app. Build is clean and the unit tests pass, but nothing covers sidebar rendering, so the manual checks under Test plan are all still open. Could you run through them, or tell me if you'd rather I do it before review?

Summary

Sleeping (dormant) sessions now respect the group filter in the sidebar. They used to be appended unconditionally at the very bottom, whatever the display mode or the active group tab.

Why

The old comment called that "kept reachable", but on a filtered tab it reads as a leak: every sleeping session from every group, listed under a tab that is supposed to show one. A sleeping session stays reachable anyway, one click away on All.

Implementation notes

Placement now follows GroupDisplayMode:

Mode	Dormant rows
FilterStrip	Still after the live rows, but only those matching ActiveGroupId. All lists every one.
InlineHeaders	At the end of their own group section, counted in that header's badge, and collapsed with the group. Ungrouped now also shows when all its members are asleep.
None	Unchanged: all at the bottom.

• AddDormantSidebarItem only registers the row now; it doesn't add it to the sidebar. RebuildSidebarOrder is the only thing that places rows, because where a row goes depends on the display mode and the filter. Every caller must rebuild afterwards. SleepSession and the wake-failure path now do (replacing a bare RefreshTerminalLayout), and the other callers already did. I checked this against main after feat(sessions): restart a session without losing it #140, and all five callers rebuild, the restart fallback included.
• Dormant rows are ordered by walking SessionManager.Sessions, not the _dormantSidebarItems dictionary, so they follow drag-reorder like live rows instead of insertion order.
• Two holes a review pass found in the first version, both fixed here:
  • Stale header after deleting a sleeping row. The ✕ removed the Border in place and never rebuilt. Now that inline headers count dormant rows, that left "Work (3)" over two rows, or an empty header. It rebuilds now.
  • A dormant row with an orphaned GroupId rendered nowhere in InlineHeaders mode, and still kept the empty-state placeholder hidden. RemoveGroup clears GroupId, so this can't happen from inside the app, but ImportExportService loads an arbitrary AppState. Such rows are now shown as ungrouped. Live sessions have the same gap and I left it alone: the proper fix is to normalize orphaned ids on load, which is a separate change.

Test plan

• dotnet build passes with 0 warnings
• dotnet test tests/CodeShellManager.Tests/ passes 597/597 (none of these cover sidebar rendering)
• FilterStrip: sleep one session in each of two groups. Each group tab shows only its own sleeping row, Ungrouped shows only ungrouped ones, and All shows every one.
• FilterStrip: sleep a session while its group's tab is active. The row appears at once; while a different tab is active, it doesn't appear.
• InlineHeaders: a sleeping row sits at the end of its own group, the header count includes it, and collapsing the group hides it.
• InlineHeaders: a group whose only members are asleep still shows its header and rows. Same for Ungrouped.
• InlineHeaders: delete a sleeping row with its ✕. The header count drops, and the header goes away (Ungrouped) or shows 0 (user group) when it was the last one.
• Wake from a filtered tab, and force a wake failure (e.g. a WSL session whose distro is gone). The row goes back into the right place.
• Drag-reorder live sessions, then sleep them. Sleeping rows keep that order.
• Restart path from feat(sessions): restart a session without losing it #140: a restart that fails falls back to a dormant row in the right group.
• --clean run: no regressions in the empty-state placeholder.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NsYm9iLc2aRnRDfV2uZRQX