unified-mcp-server: health surface for the shared daemon

[jobordu]

Part of #408. Depends on daemon lifecycle.

Why

Once one process serves every session, its health is everyone's problem. Today a broken MCP server is visible because it belongs to one session; a broken singleton is invisible until several sessions fail at once.

What

• a health endpoint reporting uptime, connected clients, per-provider reachability
• structured logging that identifies which client a failure belongs to
• surfaced in nforma tui / status output alongside existing checks

Acceptance

• health endpoint answers while providers are unreachable (it must not fail with them)
• logs attribute an error to a client/session
• status output shows daemon up/down and client count