workspace: Fix recent-projects cleanup wiping active workspaces

[transitoryangel]

The recent-projects picker, sidebar, welcome screen, and agent thread store all called recent_workspaces_on_disk, which combined listing with deleting stale rows. Its retention predicate rejected workspaces with no on-disk directory, including empty workspaces holding unsaved scratch buffers, and the resulting delete_workspace_by_id call cascaded into items, pane_groups, and the per-editor tables. For clarity, the method has been renamed to recent_workspaces_for_ui.

Meanwhile, last_session_workspace_locations used a slightly different form of the same predicate. The two disagreeing on what counts as a valid workspace caused #48799, Workspace WorkspaceId(N) not found errors on repeated launches (#50409), the test_window_edit_state_restoring_enabled flake (#50871), and the foreign key constraint fail on projects: open recent with a dirty scratch buffer (#51456).

Note that for the last issue mentioned (#51456) there is no save prompt for scratch buffers. This seems out of scope for this PR so I'll fix that after this is addressed.

Self-Review Checklist:

• I've reviewed my own diff for quality, security, and reliability
• Unsafe blocks (if any) have justifying comments
• The content is consistent with the UI/UX checklist
• Tests cover the new/changed behavior
• Performance impact has been considered and is acceptable

Closes #48799
Closes #50409
Closes #50871
Closes #51456

Release Notes:

• Fixed unsaved scratch buffers being lost across restarts and an occasional error when opening a recent project.

[maxbrunsfeld]

Thanks for the PR @transitoryangel. This seems like a good direction, and I'm glad you figured out the source of those bugs. A couple of thoughts:

I don't think the GC logic is quite right. In particular, I think there's no need to have conditional behavior depending on whether a workspace has any items (currently workspace_has_items influences the is_stale check). As written, it seems like if a workspace does have items, we will preserve it in the database indefinitely. But I don't think there is any benefit to keeping a workspace in the DB if it has no paths, and it is not part of the last session (it would never be returned from any query).

I think a better fix for the bugs you identified would be to adjust the workspace GC so that it never deletes the workspace for the last session or the current session. When you reopen Zed, we should bring back all of your previously-opened workspaces, even if they have zero items.

I think that GC should explicitly happen after we have loaded the last session's workspaces, and it should delete any workspaces that have no valid paths (regardless of items).

Also, in terms of naming, I don't think think the name recent_workspaces_for_ui is quite right; the 'for ui' thing doesn't add a lot of specificity, since most database queries ultimately serve to populate some UI, directly or indirectly. I agree that _on_disk was a questionable suffix too. What do you think about the name recent_project_workspaces, to convey that it is tied to the "recent projects" feature?

[transitoryangel]

Thanks for the feedback! The logic you suggest seems sound, and I'll work on changing things around.

Regarding the recent_workspaces_for_ui name - that was spotty in my opinion as well. recent_project_workspaces instead sounds pretty reasonable. I'll see how it fits in. Thank you for that suggestion!

[maxbrunsfeld]

Great work @transitoryangel. Thank you!