Migrate all backend-API callers from effective_api_url to effective_backend_api_url (local-AI users)

[oxoxDev]

Follow-up to PR #1630 (closed OPENHUMAN-TAURI-51, -80, -7Z).

#1630 fixed the IntegrationClient URL-base leak when users set config.api_url to a local-AI endpoint. The same class of bug exists across most non-integrations domains that still call effective_api_url. graycyrus enumerated 38+ call sites in the #1630 review:

• billing/ops.rs
• team/ops.rs
• referral/ops.rs
• webhooks/ops.rs
• credentials/ops.rs (9 call sites)
• channels/controllers/ops.rs (9 call sites)
• voice/
• socket/
• app_state/
• core/jsonrpc.rs

Plan

1. Rename effective_integrations_api_url → effective_backend_api_url (more general name; documents that it is the canonical resolver for ALL backend-proxied calls, not just integrations).
2. Migrate every effective_api_url call site that targets the OpenHuman hosted backend to effective_backend_api_url.
3. Keep effective_api_url for the OpenAI-compatible chat-completions client only (src/openhuman/providers/...) — that one legitimately honors the user's local-AI override.
4. Add a clippy lint or a CI grep gate to prevent future regressions.

Why

Users on local AI (Ollama, vLLM, LM Studio) currently silently lose every backend feature outside Composio integrations — billing checks, team membership, referral, webhooks, credentials, channels, voice. #1630 only stopped the Sentry flood for the IntegrationClient slice; the underlying UX issue remains for ~9 other domains.