fix(openclaw): avoid gateway restart when switching models

[winsan-zhang]

Summary

• Eliminate OpenClaw gateway process restarts when switching
  between 套餐模型 (lobsterai-server) and custom providers, or
  between different custom providers
• Use per-provider env vars (LOBSTER_APIKEY_) instead of a
  single shared LOBSTER_PROVIDER_API_KEY, so all apiKeys are
  pre-injected at gateway startup
• Model switching now only rewrites openclaw.json, which OpenClaw
  hot-reloads silently via its file-watcher — no process restart
  needed

Problem

When users switch models across provider types (e.g.,
lobsterai-server ↔ moonshot), the gateway was killed and restarted
because:

1. collectSecretEnvVars() returned a single
   LOBSTER_PROVIDER_API_KEY set to the active provider's apiKey
2. Switching providers changed this value → secretEnvVarsChanged =
   true
3. Since env vars are fixed at process spawn time, a full process
   restart was forced

Key changes:

• claudeSettings.ts: New resolveAllProviderApiKeys() — enumerates
  all enabled providers and returns their apiKeys keyed by stable
  env var suffix
• openclawConfigSync.ts:
  • New providerApiKeyEnvVar() helper — generates
    provider-specific env var names
  • buildProviderSelection() — all 4 provider cases now use
    ${LOBSTER_APIKEY_} instead of ${LOBSTER_PROVIDER_API_KEY}
  • collectSecretEnvVars() — collects all provider apiKeys
    upfront; legacy LOBSTER_PROVIDER_API_KEY set to fixed value

Test Plan

• Unit tests for env var naming convention and switch stability
• Manual: switch between 套餐 and 自定义 models — verify no
  stopGateway/startGateway in logs
• Manual: switch between different custom providers — verify no
  restart
• Manual: send messages after switch — verify correct model and
  apiKey used
• Manual: edit a provider's apiKey — verify gateway restarts once
  (expected)