Catch up on the events ledger: the cold-start brief, or a delta since a cursor

Same contract as MCP's catch_up (catchUpQuerySchema in lib/schemas/events.ts validates both). Without cursor: the brief — the NON-ARCHIVED boards you can read (each with its id) with their lists (with their ids) and card counts, a tail of recent activity (recent is a taste of the ledger, not a page — no hasMore — and, unlike the inventory, it does include archived boards), per workspace the care counts stale, inventory and attention (BriefWorkspace says what each counts), and a fresh cursor. With cursor: { events, cursor, hasMore, resynced? }, only what happened since. Activity on boards you cannot read is filtered out after the scan, so a page may carry fewer events than limit — even none — while hasMore is true: keep calling with the returned cursor until hasMore is false. Keep the cursor verbatim between sessions, like a cardKey; never parse it. limit (1-100, default 50) is ignored without a cursor and split evenly across the workspaces in scope (floor, minimum one each — with more workspaces than limit, a page can carry one event per workspace and exceed it). A position retention has pruned past is restarted from the oldest event that workspace still has and reported in resynced (the delta's optional array of slugs) — not an error, and the other workspaces' deltas in the same response are unaffected. Only a cursor that is damaged or names a position this ledger never issued answers 410 STALE_CURSOR: call again without one to restart from the brief.