feat: actionable guidance when api.sandbase.ai is unreachable (0.3.2)

[liyangbing]

Follow-up to #64, which made api.sandbase.ai the only API host and dropped the implicit www fallback. On networks where api.sandbase.ai is blocked (proxy, firewall, or regional routing — observed: TCP timeout to the EdgeOne IP from some CN networks), the CLI/bridge now point the user and the agent at the one self-serve escape hatch instead of a dead end:

• NETWORK error and the bridge offline message now name SANDBASE_MCP_FALLBACK_URLS (and SANDBASE_MCP_URL) alongside sandbase doctor.
• doctor's fallback line no longer claims the bridge switches automatically (it only does so when a mirror is configured).
• Skill 'unreachable' row documents the fallback escape hatch.

No endpoint/default changes — api.sandbase.ai stays the only default host. Version 0.3.2. npm test: 172 pass.

Note: api.sandbase.ai (EdgeOne 43.159.98.112) is TCP-unreachable from my test machine; please confirm it is healthy in production from the networks you care about.

[liyangbing]

Wrong approach: api.sandbase.ai is reachable (my earlier timeout was a local proxy issue). We should keep api.sandbase.ai as the single host and not steer users to fallback mirrors. Closing.