fix: keep runtime alive during channel recovery

[tylerslaton]

What changed

• If the initial 15-second Channel readiness wait expires while the runtime is still connecting or reconnecting, start the HTTP server so Railway health checks keep the process alive.
• Continue waiting for Channel readiness in the background.
• Keep permanent initial failures terminal, and shut down if background recovery later becomes terminal.

Why

A runtime that starts during a realtime gateway outage currently exits before a retry can connect after the gateway recovers. Keeping the HTTP health route alive lets the CopilotKit retry loop eventually restore the managed Channel without a process restart.

Dependency

Depends on CopilotKit/CopilotKit#6347 and a CopilotKit release that includes it. That change retries transient initial gateway activation with exponential backoff.

Validation

• pnpm exec vitest run app/server-recovery.test.ts app/server.test.ts --reporter=verbose — 6 tests passed
• git diff --check

Existing repository failures

• pnpm check-types fails unchanged on the clean base checkout because @copilotkit/channels 0.6.1 and the runtime transitive channels-core 0.6.0 expose incompatible types.
• The full test suite also has existing failures from the same package skew. The two server test files changed or exercised here pass.