feat: viewer should connect to cb serve API instead of reading files from disk

[rajnavakoti]

Description

The viewer currently reads entity files from disk at build time (via entity-loader.ts). It does NOT connect to the cb serve API. This means users who run cb serve --block my-block still see the wrong data in the viewer — it falls back to whatever entities it finds locally (usually the healthcare demo in synthetic-domains/).

The workaround is setting CB_OUTPUT_DIR=/path/to/block before starting the viewer, but this is undiscoverable and breaks the mental model that cb serve + viewer work together.

Current Behavior

1. cb serve --block cortex starts API on port 8321 — serves cortex entities correctly
2. npm run dev in viewer/ — reads files from disk, ignores the API
3. User sees healthcare demo instead of their block data

The viewer has two disconnected data paths:

• Digest/Graph/Gaps pages: read from disk via entity-loader.ts (build-time)
• Ask page: fetches from cb serve API at runtime (correct)

Acceptance Criteria

• Viewer pages (digest, graph, gaps, evals) fetch data from cb serve API at runtime, not from disk at build time
• Running cb serve --block my-block + opening the viewer shows that block's data automatically
• No CB_OUTPUT_DIR env var needed for the standard workflow
• Fallback: if no API is running, viewer shows a clear "connect to cb serve" message instead of silently loading demo data

Out of Scope

• Changing how cb serve works (API is fine)
• Viewer redesign or new pages
• SSR/deployment concerns — dev mode is the priority