feat(workspace): add workspace alert IPC

[khughitt]

What this does

Adds a workspace-alert feature. Over IPC, a user can flag a workspace — or the workspace containing a given window — so it renders with the existing urgent style until they switch to it. The alert auto-clears when the user activates that workspace.

IPC commands:

• workspace-alert-add <workspace-id>
• workspace-alert-add-window <window-id>
• workspace-alert-clear <workspace-id>
• workspace-alert-clear-all
• workspace-alert-status

Why this is a rewrite of #3025

The original PR (#3025) was declined on scope: it introduced a new canonical Workspace.key field and threaded it through all seven compositor backends — the highest-maintenance surface in the project — plus output-name-resolver plumbing and widget/snapshot changes.

This version reuses the compositor's existing workspace identifiers as the alert token, so the feature collapses to a self-contained unit:

• a WorkspaceAlertService (set of tokens + overlay logic),
• the five IPC handlers,
• one overlay hook in each CompositorPlatform::workspaces() overload,
• the focused-output auto-clear wiring.

An alert token is matched against a workspace's id, name, or visible index, because backends populate different ones (niri only sets the visible index; Sway uses the name as the id; tag compositors use the numeric id). This keeps the feature usable across compositors without threading a new canonical key through every backend. No per-backend changes. The only cross-cutting edit is relocating the plain WorkspaceWindowAssignment struct between two headers.

Known limitation

None of these identifiers is guaranteed unique across outputs on compositors that number workspaces per monitor (niri index, mango/dwl tags), so an alert there applies to the matching workspace on every monitor (and activating it on any monitor clears it). This is the accepted per-output trade-off for not introducing a per-backend canonical key; it's documented in the service header.

Notes

• The overlay reuses the urgent flag and only marks inactive rows, so alerting the active workspace simply shows once you leave it and clears on return; on per-output compositors this also lets an inactive duplicate alert even when the same workspace is active on another monitor.
• The id/name/index matching runs after the metadata backend's apply() in workspaces(), so the visible index is populated at match time (this is what makes niri — which doesn't expose a usable id — work).
• m_workspaceAlertService is declared before m_compositorPlatform so the raw pointer handed to the platform outlives it.

Testing

• just build passes.
• workspace_alert_service unit test: 31/31 (covers overlay suppression of active rows, matching by index/name/id, per-output duplicate-index handling, window→workspace resolution, and validation).
• Verified live on niri: workspace-alert-add <n> by visible number, add-window <id>, clear, clear-all, and status all behave as expected.
• git diff --check clean.