Seed/demo dataset: make demo mode show realistic dummy data

[offendingcommit]

Problem

Demo mode currently only masks real data (useDemo + maskValue). If a user has no Honcho instance, or wants to evaluate OpenConcho before connecting one, there's nothing to see.

Proposal

Two complementary tracks:

1. In-app demo mode = fixture data

Extend demo mode so when toggled on (or when no baseUrl is configured), the API layer (src/api/queries.ts) serves a curated in-memory fixture:

• 2–3 workspaces with believable names
• A handful of peers, sessions, messages, conclusions
• Realistic queue activity (some pending, some done)
• Stable IDs so screenshots are reproducible

Implementation sketch:

• Wrap client.current so when demo === true it returns a fake fetcher that resolves against fixtures keyed by URL/method.
• Fixtures live under src/lib/demo-fixtures/ as typed objects derived from components[\"schemas\"].

2. Seed script for a real Honcho instance

A pnpm seed script (or scripts/seed.ts) that POSTs the same fixtures to a configured Honcho instance via the public API. Useful for:

• Local development against a fresh Honcho.
• Reproducible Playwright e2e runs.
• Demo videos.

Why both

• Fixture-mode = zero-setup evaluation, screenshots, e2e tests.
• Seed script = exercises the real API path, catches schema drift.

Open questions

• Do we ship fixtures generated from a recorded session, or hand-author for narrative quality?
• Should demo mode be the default landing experience for first-time visitors (no config required)?