feat(llm): add custom providers and dynamic API routing

[yaukwan]

Why

The current setup flow and provider routing still assume a fixed set of legacy keys (anthropic_key, openai_key, etc.). This makes setup-mode detection brittle and limits support for custom providers, custom base URLs, and multiple API protocols.

What Changed

1) Provider model and config loading

• Added structured provider config to LlmConfig via providers: HashMap<String, ProviderConfig>.
• Added ApiType (anthropic, openai_completions, openai_responses) and ProviderConfig (api_type, base_url, api_key, name).
• Added TOML parsing support for custom providers under [llm.provider.<id>].
• Added validation for provider IDs (length/characters) and provider base_url.
• Normalized provider IDs to lowercase for stable lookup.

2) Backward compatibility + setup behavior

• Legacy keys are still supported and auto-mapped into internal provider entries (anthropic, openai, openrouter).
• Explicit custom provider definitions take precedence over auto-migrated legacy entries with the same provider ID.
• load_from_env() now builds provider entries from environment keys (same behavior as TOML loading path).
• needs_onboarding() now checks a broader set of legacy provider env keys and provider-related env signals so setup mode is more accurate.

3) Runtime routing and API integration

• Refactored LlmManager to resolve provider data through get_provider() and dynamic get_api_key().
• Refactored SpacebotModel to route by model prefix (<provider_id>/<model>) and select call path from ApiType.
• Added OpenAI Responses API path support and parser/conversion flow.
• Updated provider calls to use configured base_url instead of hardcoded endpoints.

4) API and docs updates

• Updated provider status API to return a dynamic provider map (instead of fixed provider fields), with backward-compatible legacy keys included as defaults when absent.
• Extended config docs with:
  • custom provider schema and examples
  • migration notes from legacy keys
  • supported API types and base URL guidance
  • routing examples using custom provider IDs
• Updated .gitignore (test_config.rs, .idea).

Compatibility Notes

• Existing configs using legacy keys continue to work.
• Model names without provider prefix still default to anthropic.
• New custom-provider model naming follows <provider_id>/<model>.

Validation

Added tests

• Added/expanded src/config.rs tests for:
  • ApiType and provider config deserialization
  • provider table parsing + lowercase normalization
  • legacy key -> provider auto-migration
  • precedence of explicit provider config over legacy migration
  • onboarding detection behavior
  • env-only loading path populating both legacy fields and providers

Recommended checks for reviewers

cargo test config::tests
cargo test
cargo clippy --all-targets --all-features