Workspace session-info aggregate endpoint

Problem

GET /workspace/:id/sessions is cursor-paginated and does not return a total. GET /daemon/status exposes live in-memory sessionCount only. Workspaces with many persisted sessions (for example from scheduled tasks) cannot learn the local store size without paging every session.

Proposal

Add:

GET /workspace/:id/session-info
GET /workspaces/:workspace/session-info

Response (illustrative):

{
  "active": 450,
  "archived": 30,
  "total": 480,
  "live": 2,
  "expensive": true,
  "cost": "disk_scan"
}

live is omitted for an untrusted secondary workspace because those catalog reads must not query the live bridge. If the scan reaches its safety limit or cannot classify a candidate JSONL file, the response includes "truncated": true; persisted counts are then lower bounds.

Cost model

Persisted counts reuse the existing full-directory scan pattern already used by session title search (SessionService.findSessionsByTitle / findSessionTitlesByPrefix):

1. readdir the project chats dir (and archive twin)
2. filter UUID *.jsonl
3. cap at the same file-processing safety limit
4. read only the first JSONL record for project-hash membership

No title/prompt hydration. This is O(n) on disk and must not be polled. The response always sets expensive: true and cost: "disk_scan" so clients can fail closed on hot paths. Docs call this out explicitly.

Default list pagination stays unchanged and does not compute totals. Do not reuse organized-view listAllPersistedSummaries for counts — that path hydrates full list metadata up to 50k sessions.

Capability

Always-on session_info on /capabilities, next to session_list.

Non-goals

• Cached counters / mutation-hook accounting (possible follow-up if call sites need lower latency)
• Stuffing total into every list page
• Organized-group or parent-filtered totals in v1