feat(cron): add configurable inference provider for AI tasks

[octo-patch]

Reason: The cron AI task runner is hard-wired to the claude CLI subscription with no way to select a configurable inference provider or its endpoints.

What changed

• Add src/daemon/providers.ts: an inference provider registry. It ships a MiniMax entry exposing both of its models (MiniMax-M3, 1,000,000-token context; MiniMax-M2.7, 204,800-token context) and both regional endpoints (global_en → https://api.minimax.io, cn_zh → https://api.minimaxi.com), plus a resolveProviderConfig resolver that validates the selected provider, region, and model.
• Wire runAiTask in src/daemon/cron-engine.ts to honor an optional openwolf.cron.provider selection block. When set, the claude CLI is pointed at the provider's Anthropic-compatible base URL (ANTHROPIC_BASE_URL + ANTHROPIC_AUTH_TOKEN) and asked for the selected model via --model; the API key is read from the provider's environment variable (MINIMAX_API_KEY). When unset, the existing subscription-credential behavior is unchanged.
• Add "provider": null to the cron block of the config template — a backward-compatible default that keeps current behavior until a provider is configured.
• Document cron.provider (id, region, model, key env) in docs/configuration.md.
• Add tests/providers.test.ts covering registry resolution, default and CN region/model selection, case-insensitive ids, and the unknown-provider/region/model error paths. These run under the repository's existing node --test tests/*.test.ts suite.

Checks

• npx tsc --noEmit — passes.